State of the Art Deep Learning Implementation for Multiclass Classification of Black Pepper Leaf Diseases

State of the Art Deep Learning Implementation for Multiclass Classification of Black Pepper Leaf Diseases | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article State of the Art Deep Learning Implementation for Multiclass Classification of Black Pepper Leaf Diseases Anita S Kini, Prema KV, Smitha N Pai This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3272019/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 16 Jan, 2024 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Black pepper is a medicinal plant that is extensively used in Ayurvedic medicine because of its therapeutic properties. Leaf diseases can be diagnosed at an early stage with the aid of a smart computer vision system and timely disease prevention can be targeted. The proposed work represents an intelligent transfer learning technique through state-of-the-art deep learning application to predict the presence of prominent diseases in black pepper leaves. The ImageNet dataset available online is used for training deep neural network, initially. Later, this trained network is utilized for the prediction of the developed black pepper leaf image dataset. The developed data set consist of real time leaf images, which are candidly taken from the fields and annotated under supervision of an expert. The leaf diseases considered including healthy leaves are anthracnose, slow wilt, early stage phytophthora, phytophthora and yellowing. The accuracy obtained with 0.001 learning rate ranges from 99.1–99.5% for the Inception V3, GoogleNet, SqueezeNet and Resnet18 models. This work represents improvement in agriculture and a cutting edge deep neural network method for early stage leaf disease identification and prediction. This is an approach using a deep learning to predict black pepper leaf disease. Physical sciences/Engineering/Electrical and electronic engineering Biological sciences/Computational biology and bioinformatics/Computational models Biological sciences/Computational biology and bioinformatics/Computational platforms and environments Biological sciences/Computational biology and bioinformatics/Data acquisition Biological sciences/Computational biology and bioinformatics/Data integration Biological sciences/Computational biology and bioinformatics/Data processing Biological sciences/Computational biology and bioinformatics/Image processing Biological sciences/Computational biology and bioinformatics/Machine learning Earth and environmental sciences/Environmental sciences Physical sciences/Energy science and technology Physical sciences/Engineering Physical sciences/Mathematics and computing Black pepper Convolutional neural network Deep learning Early-stage Leaf diseases Image segmentation Transfer Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Figure 17 Figure 18 1. INTRODUCTION Plant disease is thought to be responsible for billion-dollar crop losses worldwide. The growth of agriculture is impeded due to various crop diseases. Farmers anticipate diseases through visual symptoms in plants. In the most cases it is misjudged, misinterpreted, and frequently wrong. This leads to the unauthorised use of plant pesticides and chemicals. The catastrophic effect of plant disease is crop loss and no harvest in some dreadful cases. Black pepper is a medicinal plant. Black pepper is referred to as "black gold" in Ayurveda. It is the most crucial component in Indian and other cuisines and has numerous medicinal and health benefits. About 80% of people in developing countries, as well as about 3% of people in developed countries, are inclined towards using medicinal plants such as black pepper to treat their alignments [1]. This is the main ingredient in many ancient, unani, and ayurvedic medicines, which can cure many chronic diseases like diabetes, high blood pressure, the common cold, cough, etc. Black pepper plants are affected by diseases such as slow wilt, quick wilt, phytophthora, foot rot and yellowing etc. It is a native spice of south Karnataka, Tamil Nadu, Kerala, and some forest states of northeast India. The countries with the highest black pepper production, including India, are Vietnam and Indonesia. There is a growing demand and requirement for better technologies to automate disease detection. Currently, pathologists and agriculturalists rely on the traditional laboratory method and detect diseases with the naked eye. Computerized detection of plant diseases is crucial to identify and detect the primary signs of diseases so that they can be treated. The scientific community has studied plant diseases extensively, mostly concentrating on the biological properties of plant illness. Plant diseases are unavoidable and have become a big challenge [2]. Crop production scales down due to influences of pests, fungal attacks, environmental effects such as an unexpected variation in the meteorological conditions, which escalate further losses. This encourages the practise of some cutting-edge techniques so that the diseases can be premeditated in the specified area for various disease variance. As a result, there will be less labour work and less time spent on conventional methods or waiting for an expert [3]. Plant diseases are dependent on specific environmental factors and soil culture. Intelligent techniques for identifying remote areas as well as adjacent areas would furnish a well-organised yield monitoring system for disease-ridden areas. For such circumstances, hyper-spectral sensors or thermography sensors could be brought into practise to limit human intervention and unreachable far-field areas [4]. The disease is visible because the crop has a patchy distribution throughout the field. For instance, studies of tomato leaves [5], rice leaves [6], and litchi leaves [7] demonstrate how vulnerable plants are to infections [8]. The latest deep learning algorithm enhances the computational approach with precise disease identification and realistic output images. Several real time computer vision systems have been recently developed using machine learning algorithms and other computational methods. Deep learning techniques are the fastest adopted techniques that help in analysis and feature extraction of dissimilar types of data [9]. Deep learning is a very popular methodology in almost all fields. Recently, it has gained tremendous popularity in agricultural domain because of its accurate results. Most of the researchers are training their data with an artificial neural network to get well-classified results [10]. Convolutional neural network (CNN) and artificial neural network (ANN) are the most successful deep learning techniques that has been associated with various image segmentation, feature extraction, and pattern or design recognition tasks. The proposed work demonstrate implementation of stat-of -the-art deep learning in categorization of black pepper leaf diseases. The various diseases considered for the proposed work are depicted in Figure 1. The dataset includes six prominent categories of black pepper leaf diseases those are Anthracnose, Early stage phytophthora, Slow wilt, Yellowing, Phytophthora and healthy leaves [11]. There have been numerous studies on different leaf and plant disease detection techniques. To our information, this is the original attempt to use transfer learning for the prediction of black pepper leaf diseases in early stage. The image dataset used in the experiment is collected from black pepper plantations from the districts of Madikeri, Karavali, Manipal, and Udupi region in southern Karnataka, India, and used as a “benchmark dataset” for the subsequent study. The important contributions made by this work are as follows: The proposed approach uses real-time images of black pepper leaves captured by various high-end electronic equipment: a Nikon Coolpix, a Sony DSLR, and an Android phone. The benchmark dataset is created and annotated under the supervision of an agriculture scientist. The state-of-art deep neural network algorithm is used along with transfer learning. The ensemble implementation of the three SOTA algorithm give better predictions and are better than working on any single deep neural network model. This is been observed using transfer learning with Inception V3 alone and implementation of a combinative implementation of SqueezeNet, GoogleNet and ResNet -18. These DNN are the faster image classification networks. Any machine learning challenge aims to choose a single model that can most accurately forecast the desired result. Ensemble approaches consider a wide range of models and average those models to build one final model, as opposed to creating one model and hoping that this model is the best/most accurate predictor we can make. The article is arranged as follows: The motivation of the proposed work is mentioned in Section 2, related study is discussed in Section 3. Section 4, deals with materials and methods. In this section, various stages related to data acquisition, data preprocessing, image segmentation, convolutional networks, and transfer learning are discussed. Followed by Section 5 where the experimental methodology is discussed with the help of a block diagram. Section 6 includes results and discussion, and Section 7 conclusion and future work where current limitations and future work are discussed. 2. MOTIVATION At present, there is a huge surge in the automation of everything existing in the world. There have been numerous studies on different leaf and plant disease detection techniques. Algorithms for machine learning have been applied to categorize leaf diseases. But the classification result is not accurate for a small dataset. There is a dearth of standard sets of evaluation criteria for leaf disease identification and classification. According to our knowledge, this is an attempt to exercise transfer learning in black pepper leaf disease detection. This work proposes a deep neural network learning implementation with high accuracy in early leaf disease prediction. This experiment will encourage implementation of advanced technique in agriculture processes and reduce manual labor and cost of disease estimation. The data set created will also help novice farmer to learn about disease it’s variety and will be able to apply preventive measure to save the crop loss at early stage itself. Significance of the given research work: 1.Early stage leaf disease detection in black pepper. 2.Improved feature extraction technique for the black pepper leaf diseases. 3.The given research work motivates contribution towards precision agriculture. 4.Encourage research in agriculture domain and usage of machine learning for betterment of people. 3. RELATED WORK Recent progresses in deep neural networks have made it possible for researchers to significantly correct the precision of object identification and recognition techniques. The CNN models used in automatic image recognition systems are particularly useful for detecting the onset of diseases at different stages of plant development. Barbedo, J.G.A [1] discussed techniques for enhancing data that lessen the impact of the absence of image databases that can accurately depict the numerous illnesses and symptoms in a given image. But at the same time, they’re unable to replicate the majority of the practical diversity. This paper deals with specific lesions and patches instead of taking the leaf as a whole into account for this purpose. Too, E.C. et al. [2] elaborated that traditional machine learning approaches like support vector machine, multilayer perceptrons, decision trees and neural networks have historically been used to solve the challenges of autonomous disease recognition in plants. The study looks into how the quantity and diversity of datasets affect the efficacy of deep learning methods used in plant pathology. Barbedo, J.G.A. [3] elaborated that today there is a shift of technologies from traditional to convolutional neural networks, which are now the predominant approach that applies deep learning concepts. Ganatra, N., and Patel, A. [5] projected a prediction model for classifying and detecting plant leaf disease detection using machine learning and computer vision approaches. Preprocessing, segmentation, and extraction of attributes such as shape, color, texture, vein etc. are done on the raw image of a leaf. Thangaraj, [8] discussed one of the venerated deep learning techniques for accurately detecting plant disease with little to no plant image data, namely transfer learning. They used a deep learning neural network model built on transfer learning to recognize tomato leaf disease. The author has used available and real-time images of tomato plants to detect disease. Additionally, stochastic gradient descent (SGD), RMS prop optimizers, and adaptive moment estimation (Adam) are used to evaluate the performance of the suggested model. The result of the experiment demonstrated that the suggested model, which use transfer learning strategy, is efficient at automatically classifying tomato leaf diseases. Ghosal et al. [9] have worked on rice leaf, which is afflicted by a number of illnesses at different phases of its development. The authors created their own dataset of rice leaves due to a lack of available picture datasets and castoff transfer learning to create a model of deep learning. The suggested CNN architecture, which is based on VGG-16, was trained and tested using data gathered from the internet in addition to data gathered from the rice field. A total of 1509 rice leaf images were tested on 647 other images. The suggested architecture was a success. Cross-validation is advised to validate results in future work. Devaraj et al. [13] frazzled on the importance of image processing in the arenas of agriculture and disease classification. Automatic disease detection is modelled using image segmentation and feature extraction. Upadhyay, et al. [14] operated on brown spot diseases in paddy plants. The classification model based on CNN is trained and validated on the three major classes. such as healthy leaves, developed spot leaves, and early stage leaves. Saleem, et al. [18] have performed an analysis of deep learning and convolutional neural network optimizers. The work represents an appraisal of a hybrid model for plant disease detection. A total of 26 different diseases fall under 14 different plant species that were classified. The authors propose that the Xception architecture, when trained with the Adam optimizer, gives high accuracy. Fuentes, et al. [28] accentuated the importance of deep learning based on recent advancements towards object detection and recognition system accuracy. Cameras with different resolutions were used to gather real-time images of diseased tomato leaves. The authors used Single Shot Multibox (SSD), a region-based fully convolutional network, and a faster region-based convolutional neural network to complete their work. Single-shot multiboxes developed by the combination of two different region-based CNNs are discussed. 4. MATERIALS AND METHODS 4.1. IMAGE ACQUISITION In the proposed research work, an expert-annotated benchmark dataset is considered. Data were directly captured from the field with the aid of high-end electronic devices: a Nikon Coolpix, a Sony DSLR, and an Android phone. Images captured by high-resolution cameras in a variety of dimensions are transformed to a uniform size, typically 300 by 300 pixels. Since the gathered images have different sizes, it is vital to make them all the same size. A collection of 1800 distinct photos of diseased and healthy black pepper leaves were considered for testing. These pictures were taken at random for both healthy and unhealthy leaves. For each disease type, almost 300 images were collected. Data collected from various region in Karnataka, India is depicted in Fig. 2 . Figure 2 . (a)(b)(c)(d)(e)(f)(g)(h)(i) shows field scouting in the black pepper field with an expert. 4.2. IMAGE PREPROCESSING The visual spectrum is sensitive to the cameras used for regular photographs. There are problems with recording live photographs, including the absence of a distinct and detailed border, an ambiguous background, symptoms of various disorders, variations in the lighting conditions, an unexpected brightness, a dull or cluttered background, and others. These elements collectively have a significant impact on image analysis. Real photos need preprocessing in order to fully analyze them. A ground truth image that is the region of interest is obtained using the grab cut algorithm. The initial step to performing the above-stated task was to mark the region of interest. To achieve this, we have used online image processing, the GIMP tool. Where we use a bounding box to mark the image. Everything outside ROI (bounding box) is considered as background and masked as zero, which is black. Next, these images are used matlab R2023a for further preprocessing. The code is written in python, importing suitable libraries. The Gaussian mixture model (GMM) is used to filter the background and foreground. GMM learns the pixel to separate background and foreground, taking into account the color pixels and the area outside the rectangular block to label the image to provide the ground truth image, Fig. 3 . Later, the division is made by implementing grab-cut algorithm, where the clustering method is used by clustering all the forefront images to one node known as the source node and all the background pixels to another node known as the sink node [ 11 ]. These pixels are hard-labelled based on their closeness to the source node as foreground and their sink node as background. The weight among the pixels is defined by the edges defined by the rectangle block. This decides the probability of grouping pixels in each terminal. The min-cut (minimum-cut) technique is used to further partition the image graph into distinct portions using the minimum cost function. Considering the probability of the pixels being next to the source node or sink node, the image is segmented, and finally, labelled annotated image data is obtained. Sample data is shown in Fig. 3 . 4.3. IMAGE AUGMENTATION Image augmentation is a method where the images are modified into a new form of the similar image. This is done to expand the training data set. Augmentation is used to represent a given image in various versions. Such that the images are flipped, rotated, and tilted to add extra records to the given database [ 12 ]. To artificially increase the training dataset, the given data is vertically rotated on the x-axis, modified at a 90-degree angle, and rescaled as elucidated in some sample in Fig. 4 (a), (b), (c). This is achieved using python’s built-in functions while training a deep neural network. 4.4. CONVOLUTIONAL NEURAL NETWORKS Feature extraction is the main reason for using machine learning for data organization. Convolutional Neural Network (CNN) is a type of machine learning algorithm that has been widely used for image categorization in recent years. These collection of algorithms in neural networks are termed as deep learning algorithms. These algorithms are the extensions of neural network processes. The convolutional neural network is utilized to automatically train the diseases [ 13 ]. Such systems are real-time deployable. Convolutional layers in CNN are followed by a pooling layer, an intermediate layer and further layers. There is an activation function in the pooling layer and between each set of the convolutional layer, Fig. 5 . The input image was processed through a filter by the convolutional layer primarily, which modifies the pixel intensities [ 14 ]. The activation function helps in determining neurons’ states, which also sends a signal to the subsequent linked neuron in the higher layers. The convolution responses are shrunk down to a smaller dimension using the pooling layer. This layer can be used with a variety of pooling techniques, including maximum, minimum, average, and other types [ 15 ] [ 16 ]. We have implemented a CNN to classify the black pepper plant leaf images based on disease categories and implemented max pooling (maximum) for multi class classification. Inception V3 The recommended Inception V3 neural network architecture, is 22 layers deep with 27 pooling layers encompassed, depicting the stepwise execution details. It consists of a total of nine linearly stacked inception modules. Terminals of the inception module are connected to the global average pooling layer. The deep neural network is designed at the succeeding layer by replacing the last few layers, including the final layer [ 17 ]. For these layers, the learning rate is set high, to incorporate a faster learning rate into the newly formed (higher-level) layers. Inception V3 is employed to classify diseases in black pepper leaves effectively. Google Net Google net was designed and proposed by google in collaboration with numerous universities. Google net consists of many different kind of 1 X 1 convolution, average pooling to create deeper architecture. The convolutional network is used to reduce the number of parameters related to weight and bias so that small network can be build up. [ 18 ] The global average pooling used at the end” of this network, averages a feature map of 7 X 7 to 1X1. Thereby decreasing trainable parameters to 0 while improving accuracy of high class. The accuracy of google net is much more improved than its predecessors. SqueezeNet The squeezenet architecture is also designed to reduce the parameter of the original data matrix. It reduces the dimensions by using 1X1 convolution by the using the design strategy that normally squeezes the parameter and the design is termed as fire modules. Squeeze Net architecture are based on embedded system which are highly resource inhibited operations. The main filter used in this system is of size 1X1 and not 3 X3 that is used in most of the deep learning models. The fire model consists of squeeze convolutional layer which has filter of size 1 X 1 this is further expanded and feed into other filters that are expandable s this would expand to mixture of 1 X 1and 3X3 filters. Squeeze net model reduces the input module to smaller networks [ 19 ]. This small network helps in reducing the size and increasing compatibility. Squeeze net requires comparatively lesser processing time, reducing CPU inferences. Resnet-18 Resnet stands for residual network. This network architecture is formed on the basis of VGG-16 network. But the layers in resnet-18 are short circuited. Figure 8 a shows the short circuit connection between the CNN layers. The concept followed here is to skip in between layers through short circuit connections. Hence this is some time called skip connections [ 20 ]. Which are responsible for further making residual blocks. Importance is given in mapping the residual block rather than the underlying layers themselves. 4.5. TRANSFER LEARNING Transfer learning is an approach that consists of domains and tasks. The domain comprises of two components. A feature space H is defined in the domain D with marginal probability distribution P(H), where H= {h 1...., h n}, h i ∈ H . A task in given specific domain D, specified as D = { H , P(F)}, consists of two components: a labelled space G and an objective predictive function f : H◊ G. Now for a new instance h, the prediction of corresponding label f (k) is thru function f . That is task T = {G, f (k)}, is discovered from the training data consisting of pairs {h i, g i }, where h i ∈ H and g i ∈ G T= {G, P(G/H)} = {G, f (k)} When there is a known source domain D S and task to learn T S , a target D T and learning task T L transfer learning accelerate the acquisition of anticipated target function f T (.) in D T using the knowledge in D S and T S . The loss function generally represents the misclassification in predicted output to the input provided [ 21 ]. In transfer learning for classification-based optimization, the loss function is estimated as cross entropy. In machine learning cross entropy, is an approach used while the algorithms are developed to predict from the model. This is defined as given equation: $$\text{H}( \text{p},\text{q})={\sum }_{\text{x}\in \text{X} }^{}p\left(x\right) Log q\left(x\right)$$ Here in the given equation x signifies the total number of values and p(x) categorizes the probability distribution in the real world and q(x) characterizes the probability of projected value in predicted environment. Overall, defining the loss function for predicted class [ 22 ]. Transfer learning implementation details is shown in Fig. 6 , in the form of block diagram with the flow of action. 5. EXPERIMENTAL SETUP In this study, we introduce an ensembled state-of-the art deep learning method for predicting black pepper leaf diseases. The process is carried out in two phase, first phase is data preparation and second phase involve implementation of deep neural networks for diseases classification. Images are collected from the field and annotated. The annotated/labelled images are used for the experiment [ 23 ]. The dataset comprises of 1500 black pepper leaf images. Out of these, the best set of images are chosen for each disease category, i.e., for anthracnose, early stage phytophthora, phytophthora, slow wilt, yellowing and healthy leaf images. Hence, a total 600 images are labelled and saved [ 24 ]. The leaf images are chosen such that the diseases are clearly evident with high visibility on the leaf surface [ 25 ]. The insinuation of diseases in black pepper is that they are specific to the given region, and symptoms vary from one region to another. Hence, care is taken to correctly identify and diagnose the leaf diseases. Figure 7 depicts data acquisition and annotation flow. Image segmentation is done using grab- cut method. Grab cut is a very user-friendly and powerful algorithm for extracting the foreground objects from a marked region of interest. Image augmentation is necessary for the deep learning models to run effectively during the training phase [ 26 ]. Image augmentation relates to image data scaling and rotation. More information in the form of data variation enhances the performance of deep neural networks. Next, there was a need to import big data set to train deep neural networks so that these trained networks could be used for comparing and processing our developed data set. Hence, for training the CNN model, we have casted-off the most widely used image dataset, ImageNet. This repository has 10 billion unique images in its library [ 27 ] [ 28 ]. This is beneficial for training and boosting the learning rate of deep neural network at the initial layer network. This data is loaded at the initial neural network layer in order to pretrain the neurons. To prevent misinterpretation, we gathered and considered the images from a single input source (one device) during the processes of segmentation and augmentation [ 29 ]. Images are resized to 256 X 256, the size expected by the input layer of each deep learning model. Large data sets are necessary for the deep learning performances; hence data augmentation is accoutered [ 30 ] [ 31 ]. These are considered the new databases for transfer learning implementation. Figure 8 (a) represents Resnet residual block and Fig. 8 (b) represents CNN implementation. The flow of proposed conceptual methodology using transfer learning is illustrates in Fig. 9 . 5.1. ARCHITECTURE FOR PROPOSED DEEP LEARNING IMPLEMENTATION Contemporary state-of-art deep neural network are the best models to utilize for any given task. Based on its accuracy, speed or any other parameters of importance, a DNN (deep neural network) might be classified as SOTA (State-of-the-art). The majority of computer vision fields, though, involve a trade-off between these measurements. In other words, even if a DNN is extremely quick, its accuracy may not be up to mark. Sometimes we can create a model with acceptable performance metrics, but it doesn’t have the necessary latency or throughput for a variety of applications, such as picture categorization and object detection. In CNN, features are extracted from training data using the convolutional layer. In this experiment, after meticulously training and working on new benchmark data set, four different deep neural networks: Inception V3, GoogleNet, SqueezeNet, and ResNet-18 are trained as state-of-the-art model for the black pepper leaf disease classification, using Matlab R2023a software. First, the data is trained using Inception V3 individually with and without transfer learning, and the results are noted. Next an ensemble of state-of-the-art deep learning models that is GoogleNet, SqueezeNet and ResNet-18 are used with the optimizers;” adam”,” sgdm” and “rmsprop”. The hyperparameters used are data partitioning, loss functioning-batch size, initial learning rate schedule, learning rate, learn rate factor, number of epochs, validation frequency and bias learn rate factor. The activation function used is ReLU, which is the most prevalent function in a deep neural network. Our classification problem is based on multiclass classification, hence SoftMax is employed. The code is written in python, importing all the necessary python packages in MATLAB. Further these networks are fine-tuned and learned higher-level layers are replaced by new layer including max-pooling, for the benchmark dataset classification. We have used cross entropy for improved classification probability and threshold cut to avoid overfitting. The accuracy obtained through transfer learning technique is very high. The proposed data is classified into: healthy and a set of diseased images. The six-leaf disease classification include Anthracnose, Early stage phytophthora, Slow wilt, Yellowing, Phytophthora and. Performance of the ensembled state-of-the-art network is measured by confusion metrics. Not much work is done in automating early leaf disease detection in black pepper leaves. Our approach is one of a kind for black pepper leaf disease detection. The implementation of deep learning algorithm has increased the performance of the classification model both in terms of accuracy and processing time using transfer learning technique. This work represents early disease classification and prediction with state-of-the-art deep learning models. This work represents implementation of high-end algorithms towards the automation of agricultural unit for the benefit of farmers and people working in fields. This is also an attempt to contribute to society, as the results obtained can be used in precision agriculture in the future. The architecture of the proposed work is depicted in Fig. 10 . 5.2.PERFORMANCE METRIX The efficiency of the proposed black pepper leaf disease detection is measured by the confusion matrix. Which gives performance measures based on true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). The confusion matrix is represented as in Table 1 . Table 1 Confusion Matrix Confusion Matrix Predicted Class Actual Class Positive Negative Positive True Positive (TP) False Positive (FP) Negative False Negative (FN) True Negative (TN) The performance that are computed through confusion matrix are specificity, sensitivity, precision, accuracy and F1-score. Specificity: It is the performance measure, which represents the ratio of true negative to that of false positive and true negative. Specificity = {TN}/{FP + TN} Sensitivity: It is the performance measure, which represents the ratio of true positive to that of true positive and false negative. It is also equivalent to recall. Sensitivity = Recall = {TP}/{TP + FN} Precision: Precision represents the ratio of true positive to that of true positive and false positive. It is also equivalent to recall. Precision = {TP}/{TP + FP} Accuracy: This performance measure is defined as the ratio of sum of true positive and true negative to that of total sum of true positive, false positive, true negative and false negative. Accuracy = {TP + TN}/{TP + TN + FP + FN} F1-score: This performance metric is the harmonic mean of recall and precision. It is measure of the accuracy. Represented as given equation. F1 = {2*Precision*Recall} {Precision Recall} = {2*TP} {2*TP + FP + FN} 6. RESULTS AND DISCUSSION We have used MATLAB R2023a for our deep neural network implementation. The deep neural network is customized accordingly to get better performance. An ensemble of deep learning models is used for six black pepper leaf disease classification. Figure 10 represents the architecture of proposed model. Table 2 shows various hypermeters used in the experiment and its definition with values. A comparative analysis is done with other deep learning models given in various research paper is shown in Table 3 .A comparison is also shown in Table 4 with the proposed model accuracy obtained in the experiment. The stage wise steps are specified and explained through algorithm: Implementing ensemble state-of-art deep learning model for prediction of black pepper leaves diseases. A total 9 different model combination is made with three different optimizers “adam”,” sgdm” and “msprop.”. The images are resized at pre-processing stage as required by the deep model algorithm i.e. 224 ×224 ×3 and 227 × 227×3, etc. The learning rate used in the algorithm is 0.001 and 0.0001. The predicted output acquired through MATLAB tool is shown in Fig. 11 . Image augmentation and segmentation is done during preprocessing stage. This helped in having better accuracy and F1 score from0.78 to 0.98. Ensemble deep learning approach with transfer learning technique resulted in benefiting high prediction of black pepper disease classification from 65–99%. This is seen with Inception V3 implementation in both cases. Table 5 & Table 6 exhibit architecture layer details and analysis result. The accuracy graph obtained is depicted in Fig. 12 (Inception V3). The performance of the state-of-the-art deep learning model is measured through confusion matrix. The resulted confusion matrix in MATLAB tool is shown in Fig. 13 & Fig. 14 . Where the comparison is shown for training as well as testing data. Figure 13 demonstrate the accuracy graph for Inception V3 and Fig. 14 represents for rest of the models. The performance measure acquired by the three models “GoogleNet”,” SqueezeNet” and “ResNet-18” is shown in Fig. 15 and Fig. 16, which is the confusion matrix for training and test set. The values are computed through confusion matrix. The statistic is displayed in Table 7 . Five different disease classes are considered with one healthy class. Figure 17 exhibits the calculated sensitivity, specificity, precision, accuracy and F1 score for the selected deep learning models. Figure 18 depicts the percentage accuracy of the prediction made for various category of leaf class by the model, based on confusion matrix. It is also observed that performance of the deep leaning models varies with the optimizers used. GoogleNet works well with rmsprop and sgdm. SqueezeNet works better with sgdm optimizer. ResNet-18 results in high accuracy with rmsprop optimizer. The proposed modified Res-18 outperforms all other deep learning models used in the experiment. But the outcome may vary for different dataset other than ImageNet and proposed diseased Black pepper leaf dataset. The efficiency of each available high-end deep neural network algorithm depends experiment and analysis. Table 2 Hyperparameters used in the experiment S. No Hyperparameters Description Value/Parameters 1 Data partition It relates to the division made w.r.t training and testing dataset Data divided as ratio of : 1.Training = 70%&Testing 30% 2.Training = 60%&Testing 40% 2 Input image size The image size required by the model being trained 1.224 × 224 × 3 2.227× 227 × 3 3 loss function Relates to error occurred with given value to that of algorithm output 0 to 1.00 4 mini-batch size Normally the batch sizes considered are power of 2. Not recommended for dataset less < 2000 64,128256,512 etc. 5 initial learning rate It is the value when neural network starts learning. 0.001/0.0001 6 learning rate schedule A parameter to adjust networks learning to that of predefined value 1.Time decay 2.Step decay 3.Exponential decay 7 number of epochs Relates to the iteratively working/training on entire batch of data 10,25,30 8 learn rate factor It is non-negative scalar value. Where learning rate of specific parameter is compared to global learning rate 10 9 bias learn rate factor Rate at which neural network learns 10 10 validation patience It is the maximum tries the algorithm makes w..r.t epoch for performance improvement 40 11 validation frequency It is the lowest value by which the given model is to validated 10 12 performance metrics Measure of the classification performance of machine learning algorithms Confusion Metrix 13 Optimizers A function that familiarizes the neural network’s weights and learning rate Adam, Sgdm, Rmsprop Table 3 Comparison result of other deep learning models Reference Classification Model Type of Leaf Accuracy Ganatra, N. and Patel, A [ 7 ] Random Forest Multiple 73.38% Saleem MH [ 8 ] Xception architecture Multiple 91.86 Shreya Ghosal [ 9 ] VGG-16 Rice Leaf 92.46% Hassan, S k Mahmudul [ 11 ] InceptionResNetV2 Plant Village Dataset 97.02% Arvind Krishnaswamy R [ 13 ] VGG-16 Selected Leaf 90. Santosh Kumar Up [ 16 ] CNN Paddy leaf 97.2% Saleem, Muhammad H [ 20 ] Deep Neural Network Tomato Leaf 95.76% Geetharamani, G [ 24 ] Deep CNN Plant Leaf Dataset 96.46% Table 4 Comparison of state of art deep learning approach for black pepper leaves Deep Neural Network Classification Model Leaf Type Accuracy Inception V3 Inception V3 (without transfer learning) Black Pepper Leaf 62.3% Inception V3 Inception V3 (with transfer learning) Black Pepper Leaf 94.6% GoogleNet GoogleNet (with transfer learning) Black Pepper Leaf 99.5% SqueezeNet SqueezeNet (with transfer learning) Black Pepper Leaf 99.4% Proposed Model ResNet-18 (with transfer learning) Black Pepper Leaf 99.67% Table 5 Inception V3 detailed parameter and architecture layer Type Patch size/stride Output size depth #1 × 1 #3 ×3 reduce #3 × 3 # 5 × 5 reduce #5 × 5 Pool proj params ops convolution 7× 7/2 112× 112× 64 1 2.7k 34M max pool 3× 3/2 56× 56× 64 0 convolution 3× 3/1 56× 56× 192 2 64 192 112K 360M max pool 3× 3/2 28× 28× 192 0 inception(3a) 28× 28× 256 2 64 96 128 16 32 32 159K 128M Inception(3b) 28× 28× 480 2 128 128 192 32 96 64 380K 304M max pool 3× 3/2 14× 14× 480 0 inception(4a) 14× 14× 512 2 192 96 208 16 48 64 364K 73M inception(4b) 14× 14× 512 2 160 112 224 24 64 64 437K 88M inception(4c) 14× 14× 512 2 128 128 256 24 64 64 463K 100M inception(4d) 14× 14× 528 2 112 144 288 32 64 64 580K 119M inception(4e) 14× 14× 832 2 256 160 320 32 128 128 840K 170M max pool 3× 3/2 7× 7× 832 0 inception(5a) 7× 7× 832 2 256 160 320 32 128 128 1072K 54M inception(5b) 7× 7× 1024 2 384 192 384 48 128 128 1388K 71M avg pool 7× 7/1 1× 1× 1024 0 dropout(40%) 1× 1× 1024 0 linear 1× 1× 1000 1 1000K 1M softmax 1× 1× 1000 0 Table 6 Inception V3 Analysis Result Name Activations Type Learnables Inception_5b_1×1 384 1×1× 832 convolutions with stride [1 1] and padding [0 0 0 0] 7 ×7 × 384 Convolution Weights 1 ×1× 832 ×384 Bias 1 ×1 ×384 Inception_5b-relu_1×1 ReLU 7 ×7 × 384 ReLU - Inception_5b-5×5_reduce 48 1×1×832 convolutions with stride [1 1] and padding [0 0 0 0] 7 ×7 × 48 convolution Weights 1 ×1× 832 ×48 Bias 1 ×1 ×48 Inception_5b-relu_5×5_reduce ReLU 7 ×7 × 48 ReLU -- Inception_5b-relu_5×5_reduce 128 5×5× 48 convolutions with stride [1 1] and padding [2 2 2 2] 7 ×7 × 128 Convolution Weights 5 ×5× 48 × 128 Bias 1 ×1 × 128 Inception_5b-relu_5×5 ReLU 7 ×7 × 128 ReLU -- Inception_5b-relu_3×3_reduce ReLU 7 ×7 × 192 ReLU -- Inception_5b-3×3 384 3×3× 192 convolutions with stride [1 1] and padding [1 1 1 1] 7 ×7 × 384 Convolution Weights 3 ×3× 192 × 384 Bias 1 ×1 × 384 Table 7 Statistic Measure for state -of -the-art deep learning model Measure sensitivity specificity precision Accuracy F1 Score Derivations TPR = TP / (TP + FN) SPC = TN / (FP + TN) PPV = TP / (TP + FP) ACC = (TP + TN) / (P + N) F1 = 2TP / (2TP + FP + FN) Predicted Classification Anthracnose (ANT) 0.9565 0.9894 0.9429 0.9843 0.9496 EarlyStagePhytophthora (ESP) 0.9753 1.0000 1.0000 0.9959 0.9875 Healthy (HLT) 0.9277 1.0000 1.0000 0.9862 0.9625 Phytophthora (PHY) 0.9178 0.9841 0.9178 0.9734 0.9178 Slow Wilt (SW) 0.8649 0.9812 0.9014 0.9620 0.8828 Yellowing(Y) 0.9552 0.9658 0.8312 0.9642 0.8889 7. CONCLUSION AND FUTURE WORK This research focuses on the early stage diagnosis of black pepper plant leaf diseases with the comprehension of computer vision. This is the first attempt of its kind to detect black pepper leaf diseases using a deep learning approach. The proposed technique helps in predicting early symptoms of various black pepper leaf diseases. For this study, five types of black pepper leaf diseases were considered: anthracnose, early stage phytophthora, quick wilt, yellowing, and slow wilt, as well as healthy leaves. The state-of-the-art deep neural networks are implemented successfully on the newly created leaf dataset with an accuracy of 99.7%. The accuracy alters with respect to optimizers used. Deep learning models work well when data is available in millions. ResNet-18 provided high accuracy 99.7% when compared with other models. This work proposes a solution towards agricultural intelligence for predicting diseases remotely with the application of high potential deep neural network in precision agriculture. The technique can be further developed as an app for smart phones. An automated leaves prediction system can also be used by non-botanical experts to quickly identify early plants diseases quite effortlessly. Deep learning can aid the task of remotely sensing fields where human intervention is vulnerable. In the majority of circumstances, the traditional approach to diagnosis disease is similar. This may lead to misinterpretation about the diseases if it is unknown. Which may further tip to uneven application of pesticides in the fields, resulting in crop loss. Though diseases can be predicted based on climate conditions, their symptoms cannot be judged. With the help of the proposed method early disease predictions is possible. This will aid in reduction of unwanted pesticides distribution. A web-based or mobile-based computer system for the automatic classification of medicinal plants such as black pepper will help the local population to improve their knowledge of medicinal plants. In the future, region-growing segmentation techniques can be used interactively for profound disease spot detection in leaves. Bountiful advanced deep learning technologies could be explored, such as VGG-101, VGG-S, U-Net, V-Net, Yolo, and Fuzzy Logic, for leaf disease detection, classification and prediction. Zone-based or global feature-based feature extraction techniques could also be explored. Synthetically developed image data can also be used in future, provided the required input data is in few hundreds. Declarations Competing interest: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper Authors Contribution Statement Anita S Kini wrote the main manuscript text, including conceptualization, Data collection and Formal Analysis, Programming and Validation.Supervision for writing is done by Prema K V & Smitha N Pai.All authors have reviewed the manuscript. Ethical & Informed consent for the data used The data used in the experiment were all obtained by the first author through field scouting. Data availability and access The data used to support the given experiment are available on request through the first author’s Email. References Barbedo, J.G.A., Plant disease identification from individual lesions and spots using deep learning. Biosystems Engineering, 180, 2019, pp.96-107. Too, E.C., Yujian, L., Njuki, S. and Yingchun, L., A comparative study of fine-tuning deep learning models for plant disease identification. Computers and Electronics in Agriculture, 161, 2019, pp.272-279. Barbedo, J.G.A. Impact of dataset size and variety on the effectiveness of deep learning and transfer learning for plant disease classification. Computers and electronics in agriculture, 153, 2018. , pp.46-53. Hossain, M.A. and Sajib, M.S.A. Classification of image using convolutional neural network (CNN). Global Journal of Computer Science and Technology. 2019. Thangaraj, Rajasekaran, S. Anandamurugan, and Vishnu Kumar Kaliappan. "Automated tomato leaf disease classification using transfer learning-based deep convolution neural network." Journal of Plant Diseases and Protection 128, no. 1 (2021): 73-86. Ghosal, Shreya, and Kamal Sarkar. "Rice leaf diseases classification using CNN with transfer learning." In 2020 IEEE Calcutta Conference (CALCON), pp. 230-236. IEEE, 2020. Ganatra, N. and Patel, A., 2020. A multiclass plant leaf disease detection using image processing and machine learning techniques. Int. J. Emerg. Technol, 11(2), pp.1082-1086. Saleem MH, Potgieter J, Arif KM. Plant disease classification: A comparative evaluation of convolutional neural networks and deep learning optimizers. Plants. 2020 Oct;9(10):1319. Sagar, Abhinav, and Dheeba Jacob. "On using transfer learning for plant disease detection." bioRxiv (2021): 2020-05. Hasan, Reem Ibrahim, Suhaila Mohd Yusuf, and Laith Alzubaidi. "Review of the state of the art of deep learning for plant diseases: a broad analysis and discussion." Plants 9, no. 10 (2020): 1302. Hassan, Sk Mahmudul, Arnab Kumar Maji, Michał Jasiński, Zbigniew Leonowicz, and Elżbieta Jasińska. "Identification of plant-leaf diseases using CNN and transfer-learning approach." Electronics 10, no. 12 (2021): 1388. Devaraj, Abirami, Karunya Rathan, Sarvepalli Jaahnavi, and K. Indira. "Identification of plant disease using image processing technique." In 2019 International Conference on Communication and Signal Processing (ICCSP), pp. 0749-0753. IEEE, 2019. Aravind, Krishnaswamy Rangarajan, and Purushothaman Raja. "Automated disease classification in (Selected) agricultural crops using transfer learning." Automatika 61, no. 2 (2020). A. S. Kini, P. K. V and S. N. Pai, "Leaves Classification Through Image Segmentation and Convolutional Neural Network," 2021 3rd International Conference on Advances in Computing, Communication Control and Networking (ICAC3N ) , Greater Noida, India, 2021, pp. 631-634, doi: 10.1109/ICAC3N53548.2021.9725744. A. S. Kini, K. Prema and S. N. Pai, "Intelligent classification model for Indian chickpea," 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 2021, pp. 1847-1852, doi: 10.1109/ICICCS51141.2021.9432129. Upadhyay, Santosh Kumar, and Avadhesh Kumar. "Early-Stage Brown Spot Disease Recognition in Paddy Using Image Processing and Deep Learning Techniques." Traitement du Signal 38, no. 6 (2021): 1755-1766. Hasan, Mosin, Bhavesh Tanawala, and Krina J. Patel. "Deep learning precision farming: Tomato leaf disease detection by transfer learning." In Proceedings of 2nd international conference on advanced computing and software engineering (ICACSE). 2019. Li, Lili, Shujuan Zhang, and Bin Wang. "Plant disease detection and classification by deep learning—a review." IEEE Access 9 (2021): 56683-56698. Arsenovic, Marko, Mirjana Karanovic, Srdjan Sladojevic, Andras Anderla, and Darko Stefanovic. "Solving current limitations of deep learning-based approaches for plant disease detection." Symmetry 11, no. 7 (2019): 939. Saleem, Muhammad Hammad, Johan Potgieter, and Khalid Mahmood Arif. "Plant disease detection and classification by deep learning." Plants 8, no. 11 (2019): 468. Saleem, Muhammad Hammad, Johan Potgieter, and Khalid Mahmood Arif. "Plant disease classification: A comparative evaluation of convolutional neural networks and deep learning optimizers." Plants 9, no. 10 (2020): 1319. Paymode, Ananda S., and Vandana B. Malode. "Transfer learning for multi-crop leaf disease image classification using convolutional neural networks VGG." Artificial Intelligence in Agriculture (2022). Kaya, Aydin, Ali Seydi Keceli, Cagatay Catal, Hamdi Yalin Yalic, Huseyin Temucin, and Bedir Tekinerdogan. "Analysis of transfer learning for deep neural network-based plant classification models." Computers and electronics in agriculture 158 (2019): 20-29. Geetharamani, G., and Arun Pandian. "Identification of plant leaf diseases using a nine-layer deep convolutional neural network." Computers & Electrical Engineering 76 (2019): 323-338. Barth, Ruud, Joris IJsselmuiden, Jochen Hemming, and Eldert J. Van Henten. "Synthetic bootstrapping of convolutional neural networks for semantic plant part segmentation." Computers and Electronics in Agriculture 161 (2019): 291-304. Kini, A.S., Reddy, P.K. and Pai, S.N., 2023. Techniques of deep learning and image processing in plant leaf disease detection: a review. International Journal of Electrical and Computer Engineering, 13(3), pp.3029-3040 . Rajesh, B., M. Vishnu Sai Vardhan, and L. Sujihelen. "Leaf Disease Detection and Classification by Decision Tree." In 2020 4th International Conference on Trends in Electronics and Informatics (ICOEI) (48184), pp. 705-708. IEEE, 2020. Coulibaly, Solemane, Bernard Kamsu-Foguem, Dantouma Kamissoko, and Daouda Traore. "Deep neural networks with transfer learning in millet crop images." Computers in Industry 108 (2019): 115-120. Vilasini, M. "The CNN Approaches for Classification of Indian Leaf Species Using Smartphones." Computers, Materials & Continua 62, no. 3 (2020): 1445-1472. Liu, Jun, and Xuewei Wang. "Plant diseases and pests detection based on deep learning: a review." Plant Methods 17, no.1 (2021): 1-18. Fuentes, Alvaro, Sook Yoon, Sang Cheol Kim, and Dong Sun Park. "A robust deep-learning-based detector for real-time tomato plant diseases and pest’s recognition." Sensors 17, no. 9 (2017): 2022. Yang, Kunlong, Weizhen Zhong, and Fengguo Li. "Leaf segmentation and classification with a complicated background using deep learning." Agronomy 10, no. 11 (2020): 1721. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 16 Jan, 2024 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 16 Nov, 2023 Reviews received at journal 07 Nov, 2023 Reviewers agreed at journal 18 Oct, 2023 Reviews received at journal 24 Sep, 2023 Reviewers agreed at journal 15 Sep, 2023 Reviewers invited by journal 15 Sep, 2023 Editor assigned by journal 15 Sep, 2023 Editor invited by journal 22 Aug, 2023 Submission checks completed at journal 22 Aug, 2023 First submitted to journal 17 Aug, 2023 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3272019","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":228261621,"identity":"01ddc2dd-d5b0-4347-87e7-b3cabc6ec07c","order_by":0,"name":"Anita S Kini","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAUlEQVRIiWNgGAWjYFACxgYgkcDDwMx8AMiQYGBgB1I8+LU0NkC0sCVAtDAT1AK2BqSaxwDCJ6SFf3Zz+4MfDGky5uw8Xzf83GGRx9/MwPjgbRtuLRJ3DjY29jDk8Fg282672XtGoljiMAOz4Vw8WhhuJDY28DBU8Bgc5t12g7dNIrHhMAObNC8eLfJALY1/wFp4nt38C9Qy/zAD+298WgyAWpp5gA4DamG7DbJlA9AWZnxaDIFaZssYpAG1sJndlm2TKDY8zNgsOeccbi1yN9IffHxTkWxvcP7ws5tv2+ry5I43H/zwpgyP9yHOQzAToOmBBJBAmvJRMApGwSgYCQAA8D1QtesXIN4AAAAASUVORK5CYII=","orcid":"","institution":"Manipal Academy of Higher Education","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Anita","middleName":"S","lastName":"Kini","suffix":""},{"id":228261622,"identity":"0643fc07-eead-4dc7-b0b4-72bd1bf93761","order_by":1,"name":"Prema KV","email":"","orcid":"","institution":"Manipal Academy of Higher Education","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Prema","middleName":"","lastName":"KV","suffix":""},{"id":228261623,"identity":"d211f7ce-3387-49d4-b4f7-20b1cbc0ac36","order_by":2,"name":"Smitha N Pai","email":"","orcid":"","institution":"Manipal Academy of Higher Education","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Smitha","middleName":"N","lastName":"Pai","suffix":""}],"badges":[],"createdAt":"2023-08-17 11:29:22","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3272019/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3272019/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-51884-0","type":"published","date":"2024-01-16T15:01:09+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":42162540,"identity":"93f993dc-0d75-4554-bb40-81cda78357f0","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1772681,"visible":true,"origin":"","legend":"\u003cp\u003eBlack Pepper Leaf Disease Dataset\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/b47df7cf916d1d91a78e59e7.png"},{"id":42162539,"identity":"94cdeea5-9536-4629-ad6b-669f79579df3","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1540963,"visible":true,"origin":"","legend":"\u003cp\u003e(a)(b)(c)(d)(e)(f)(g)(h)(i) Field scouting in the black pepper field with an expert.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/99e46b90062a05e935065c41.png"},{"id":42163446,"identity":"af826d9b-e85f-40e6-917f-e749f0a71c9f","added_by":"auto","created_at":"2023-08-25 19:21:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1377747,"visible":true,"origin":"","legend":"\u003cp\u003eSample Annotated (ground truth) Images\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/310dc0aadcbb711962d55694.png"},{"id":42163536,"identity":"3be33737-9f3d-4f67-b146-44ab0e0caedd","added_by":"auto","created_at":"2023-08-25 19:29:36","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":742849,"visible":true,"origin":"","legend":"\u003cp\u003eImage Augmentation (a) Random reflection x axis\u003c/p\u003e\n\u003cp\u003e(b)Random rotation (c) Random rescaling\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/dfebd0aad1fce1aa24002e25.png"},{"id":42162546,"identity":"1a294f6c-1fc8-4afa-b831-3872081d03c3","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":182405,"visible":true,"origin":"","legend":"\u003cp\u003eCNN Architecture\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/acff02a2131885bb5e3cea55.png"},{"id":42163444,"identity":"a07244e5-961a-46cf-825e-07287de47260","added_by":"auto","created_at":"2023-08-25 19:21:36","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":15228,"visible":true,"origin":"","legend":"\u003cp\u003eTransfer Learning Block Diagram\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/5808be7a89fcadd73182317a.png"},{"id":42162538,"identity":"d669b118-d709-45af-b167-7a3507a22d32","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":16615,"visible":true,"origin":"","legend":"\u003cp\u003eData Acquisition and Annotation\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/63f062e82f96341b3a2e8e1e.png"},{"id":42163158,"identity":"9598d4e7-db4d-4489-9a7b-68db4c4a8514","added_by":"auto","created_at":"2023-08-25 19:13:36","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":335262,"visible":true,"origin":"","legend":"\u003cp\u003e(a). Resnet residual block [14]\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e(b). CNN Implementation\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/c33b1f69334918cbeae82849.png"},{"id":42162552,"identity":"a41ae72d-ab45-4f9c-a748-1f16e26050f2","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":178420,"visible":true,"origin":"","legend":"\u003cp\u003eProposed Conceptual Model\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/266622f6edf89ad79df9fb6a.png"},{"id":42162544,"identity":"faf6f93e-a10c-4520-83c1-73e34a3d20e4","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":278830,"visible":true,"origin":"","legend":"\u003cp\u003eArchitecture of Proposed Model\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/88da481ab280a7828e0d31f8.png"},{"id":42162550,"identity":"daf13dd8-1546-4503-8f07-e9c31a470e4d","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":412792,"visible":true,"origin":"","legend":"\u003cp\u003ePredicted Black Pepper Leaves Diseases (a) Input, (b) Output.\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/c5c023a1bef5a1bb57fb0db0.png"},{"id":42163161,"identity":"8e575eb3-d73b-49ea-a894-aede3df4216d","added_by":"auto","created_at":"2023-08-25 19:13:36","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":40456,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Measure of Inception V3\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/076e511ea0fcdb2dffb94b3a.png"},{"id":42163164,"identity":"e5a461d4-97cf-49bc-ab4b-bcfdfe4cb18d","added_by":"auto","created_at":"2023-08-25 19:13:36","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":254235,"visible":true,"origin":"","legend":"\u003cp\u003eAccuracy and Loss graph Inception V3 implementation\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/9e7c3304469ab561abd13c26.png"},{"id":42162554,"identity":"2b7933ff-00ee-4a97-b8a0-2637d56f7434","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":133940,"visible":true,"origin":"","legend":"\u003cp\u003eAccuracy and loss graph for the black pepper leaves disease prediction\u003c/p\u003e","description":"","filename":"14.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/eac0d3bd5b5d82d79d2a37a3.png"},{"id":42162548,"identity":"a9842046-23d6-4a71-ac67-2a99a2a12343","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":15,"title":"Figure 15","display":"","copyAsset":false,"role":"figure","size":106467,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix (Black pepper leaf disease, state-or-the-art implementation).\u003c/p\u003e","description":"","filename":"15.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/2905e4fef438cdf016f4f043.png"},{"id":42162555,"identity":"1d911c38-15d1-4795-a7d7-2dbedf94f8ff","added_by":"auto","created_at":"2023-08-25 19:05:36","extension":"png","order_by":16,"title":"Figure 16","display":"","copyAsset":false,"role":"figure","size":128693,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion Matrix (Black pepper leaves).\u003c/p\u003e","description":"","filename":"16.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/24c36279ade74bcd156f19b5.png"},{"id":42163165,"identity":"9937366f-fd14-438d-b365-591809d37275","added_by":"auto","created_at":"2023-08-25 19:13:36","extension":"png","order_by":17,"title":"Figure 17","display":"","copyAsset":false,"role":"figure","size":51716,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Measure of state-of-the-art deep neural network\u003c/p\u003e","description":"","filename":"17.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/dbda453e2e9f44400a800948.png"},{"id":42163162,"identity":"bab8e078-ecec-451c-9530-7d3e4b3c0096","added_by":"auto","created_at":"2023-08-25 19:13:36","extension":"png","order_by":18,"title":"Figure 18","display":"","copyAsset":false,"role":"figure","size":25169,"visible":true,"origin":"","legend":"\u003cp\u003eAccuracy in percentage (%) of predicted black pepper leaf diseases\u003c/p\u003e","description":"","filename":"18.png","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/f865a45a38326620e8da2c50.png"},{"id":49978839,"identity":"31a2e735-d7f7-4cea-9904-e29fab64f34f","added_by":"auto","created_at":"2024-01-22 15:09:31","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":7492321,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3272019/v1/e709983e-40b6-42dd-a7be-93dd595f4652.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"State of the Art Deep Learning Implementation for Multiclass Classification of Black Pepper Leaf Diseases","fulltext":[{"header":"1. INTRODUCTION","content":"\u003cp\u003ePlant disease is thought to be responsible for billion-dollar crop losses worldwide. The growth of agriculture is impeded due to various crop diseases. Farmers anticipate diseases through visual symptoms in plants. In the most cases it is misjudged, misinterpreted, and frequently wrong. This leads to the unauthorised use of plant pesticides and chemicals. The catastrophic effect of plant disease is crop loss and no harvest in some dreadful cases. Black pepper is a medicinal plant. Black pepper is referred to as \u0026quot;black gold\u0026quot; in Ayurveda. It is the most crucial component in Indian and other cuisines and has numerous medicinal and health benefits. About 80% of people in developing countries, as well as about 3% of people in developed countries, are inclined towards using medicinal plants such as black pepper to treat their alignments [1]. This is the main ingredient in many ancient, unani, and ayurvedic medicines, which can cure many chronic diseases like diabetes, high blood pressure, the common cold, cough, etc. Black pepper plants are affected by diseases such as slow wilt, quick wilt, phytophthora, foot rot and yellowing etc. It is a native spice of south Karnataka, Tamil Nadu, Kerala, and some forest states of northeast India. The countries with the highest black pepper production, including India, are Vietnam and Indonesia. There is a growing demand and requirement for better technologies to automate disease detection. Currently, pathologists and agriculturalists rely on the traditional laboratory method and detect diseases with the naked eye. Computerized detection of plant diseases is crucial to identify and detect the primary signs of diseases so that they can be treated. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe scientific community has studied plant diseases extensively, mostly concentrating on the biological properties of plant illness. Plant diseases are unavoidable and have become a big challenge [2]. Crop production scales down due to influences of pests, fungal attacks, environmental effects such as an unexpected variation in the meteorological conditions, which escalate further losses. This encourages the practise of some cutting-edge techniques so that the diseases can be premeditated in the specified area for various disease variance. As a result, there will be less labour work and less time spent on conventional methods or waiting for an expert [3]. Plant diseases are dependent on specific environmental factors and soil culture. Intelligent techniques for identifying remote areas as well as adjacent areas would furnish a well-organised yield monitoring system for disease-ridden areas. For such circumstances, hyper-spectral sensors or thermography sensors could be brought into practise to limit human intervention and unreachable far-field areas [4]. The disease is visible because the crop has a patchy distribution throughout the field. For instance, studies of tomato leaves [5], rice leaves [6], and litchi leaves [7] demonstrate how vulnerable plants are to infections [8]. The latest deep learning algorithm enhances the computational approach with precise disease identification and realistic output images.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSeveral real time computer vision systems have been recently developed using machine learning algorithms and other computational methods. Deep learning techniques are the fastest adopted techniques that help in analysis and feature extraction of dissimilar types of data [9]. Deep learning is a very popular methodology in almost all fields. Recently, it has gained tremendous popularity in agricultural domain because of its accurate results. Most of the researchers are training their data with an artificial neural network to get well-classified results [10]. Convolutional neural network (CNN) and artificial neural network (ANN) are the most successful deep learning techniques that has been associated with various image segmentation, feature extraction, and pattern or design recognition tasks.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;The proposed work demonstrate implementation of stat-of -the-art deep learning in categorization of black pepper leaf diseases. The various diseases considered for the proposed work are depicted in Figure 1. The dataset includes six prominent categories of black pepper leaf diseases those are Anthracnose, Early stage phytophthora, Slow wilt, Yellowing, Phytophthora and healthy leaves [11]. There have been numerous studies on different leaf and plant disease detection techniques. To our information, this is the original attempt to use transfer learning for the prediction of black pepper leaf diseases in early stage. The image dataset used in the experiment is collected from black pepper plantations from the districts of Madikeri, Karavali, Manipal, and Udupi region in southern Karnataka, India, and used as a \u0026ldquo;benchmark dataset\u0026rdquo; for the subsequent study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe important contributions made by this work are as follows:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eThe proposed approach uses real-time images of black pepper leaves captured by various high-end electronic equipment: a Nikon Coolpix, a Sony DSLR, and an Android phone.\u003c/li\u003e\n \u003cli\u003eThe benchmark dataset is created and annotated under the supervision of an agriculture scientist.\u003c/li\u003e\n \u003cli\u003eThe state-of-art deep neural network algorithm is used along with transfer learning.\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe ensemble implementation of the three SOTA algorithm give better predictions and are better than working on any single deep neural network model. This is been observed using transfer learning with Inception V3 alone and implementation of a combinative implementation of SqueezeNet, GoogleNet and ResNet -18. These DNN are the faster image classification networks. Any\u0026nbsp;machine\u0026nbsp;learning\u0026nbsp;challenge\u0026nbsp;aims\u0026nbsp;to\u0026nbsp;choose\u0026nbsp;a\u0026nbsp;single\u0026nbsp;model\u003c/p\u003e\n\u003cp\u003ethat can most accurately forecast the desired result. Ensemble approaches consider a wide range of models and average those models to build one final model, as opposed to creating one model and hoping that this model is \u0026nbsp;the best/most accurate predictor we can make.\u003c/p\u003e\n\u003cp\u003eThe article is arranged as follows: The motivation of the proposed work is mentioned in Section 2, related study is discussed in Section 3. Section 4, deals with materials and methods. In this section, various stages related to data acquisition, data preprocessing, image segmentation, convolutional networks, and transfer learning are discussed. Followed by Section 5 where the experimental methodology is discussed with the help of a block diagram. Section 6 includes results and discussion, and Section 7 conclusion and future work where current limitations and future work are discussed.\u003c/p\u003e"},{"header":"2. MOTIVATION","content":"\u003cp\u003eAt present, there is a huge surge in the automation of everything existing in the world. There have been numerous studies on different leaf and plant disease detection techniques. Algorithms for\u0026nbsp;machine\u0026nbsp;learning\u0026nbsp;have\u003c/p\u003e\n\u003cp\u003ebeen\u0026nbsp;applied\u0026nbsp;to\u0026nbsp;categorize\u0026nbsp;leaf\u0026nbsp;diseases. But the classification result is not accurate for a small dataset. There\u0026nbsp;is a dearth of standard sets of evaluation criteria for leaf disease identification and classification. According to our knowledge, this is an attempt to exercise transfer learning in black pepper leaf disease detection. This work proposes a deep neural network learning implementation with high accuracy in early leaf disease prediction. This experiment will encourage implementation of advanced technique in agriculture processes and reduce manual labor and cost of disease estimation. The data set created will also help novice farmer to learn about disease it\u0026rsquo;s variety and will be able to apply preventive measure to save the crop loss at early stage itself.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Significance of the given research work:\u003c/p\u003e\n\u003cp\u003e1.Early stage leaf disease detection in black pepper.\u003c/p\u003e\n\u003cp\u003e2.Improved feature extraction technique for the black pepper leaf diseases.\u003c/p\u003e\n\u003cp\u003e3.The given research work motivates contribution towards precision agriculture.\u003c/p\u003e\n\u003cp\u003e4.Encourage research in agriculture domain and usage of machine learning for betterment of people.\u003c/p\u003e"},{"header":"3. RELATED WORK ","content":"\u003cp\u003eRecent progresses in deep neural networks have made it possible for researchers to significantly correct the precision of object identification and recognition techniques. The CNN models used in automatic image recognition systems are particularly useful for detecting the onset of diseases at different stages of plant development.\u003c/p\u003e\n\u003cp\u003eBarbedo, J.G.A [1] discussed techniques for enhancing data that lessen the impact of the absence of image databases that can accurately depict the numerous illnesses and symptoms in a given image. But at the same time, they\u0026rsquo;re unable to replicate the majority of the practical diversity. This paper deals with specific lesions and patches instead of taking the leaf as a whole into account for this purpose.\u003c/p\u003e\n\u003cp\u003eToo, E.C. et al. [2] elaborated that traditional machine learning approaches like support vector machine, multilayer perceptrons, decision trees and neural networks have historically been used to solve the challenges of autonomous disease recognition in plants. The study looks into how the quantity and diversity of datasets affect the efficacy of deep learning methods used in plant pathology.\u0026nbsp;Barbedo, J.G.A. [3] elaborated that today there is a shift of technologies from traditional to convolutional neural networks, which are now the predominant approach that applies deep learning concepts.\u003c/p\u003e\n\u003cp\u003eGanatra, N., and Patel, A. [5] projected a prediction model for classifying and detecting plant leaf disease detection using machine learning and computer vision approaches. Preprocessing, segmentation, and extraction of attributes such as shape, color, texture, vein etc. are done on the raw image of a leaf. Thangaraj, [8] discussed one of the venerated deep learning techniques for accurately detecting plant disease with little to no plant image data, namely transfer learning. They used a deep learning neural network model built on transfer learning to recognize tomato leaf disease. The author has used available and real-time images of tomato plants to detect disease. Additionally, stochastic gradient descent (SGD), RMS prop optimizers, and adaptive moment estimation (Adam) are used to evaluate the performance of the suggested model. The result of the experiment demonstrated that the suggested model, which use transfer learning strategy, is efficient at automatically classifying tomato leaf diseases.\u003c/p\u003e\n\u003cp\u003eGhosal\u003cem\u003e\u0026nbsp;\u003c/em\u003eet al. [9] have worked on rice leaf, which is afflicted by a number of illnesses at different phases of its development. The authors created their own dataset of rice leaves due to a lack of available picture datasets and castoff transfer learning to create a model of deep learning. The suggested CNN architecture, which is based on VGG-16, was trained and tested using data gathered from the internet in addition to data gathered from the rice field.\u0026nbsp;A total of 1509 rice leaf images were tested on 647 other images. The suggested architecture was a success. Cross-validation is advised to validate results in future work. Devaraj et al. [13] frazzled on the importance of image processing in the arenas of agriculture and disease classification. Automatic disease detection is modelled using image segmentation and feature extraction. Upadhyay, et al. [14] operated on brown spot diseases in paddy plants. The classification model based on CNN is trained and validated on the three major classes. such as healthy leaves, developed spot leaves, and early stage leaves.\u003c/p\u003e\n\u003cp\u003eSaleem, et al. [18] have performed an analysis of deep learning and convolutional neural network optimizers. The work represents an appraisal of a hybrid model for plant disease detection. A total of 26 different diseases fall under 14 different plant species that were classified. The authors propose that the Xception architecture, when trained with the Adam optimizer, gives high accuracy. Fuentes, et al. [28] accentuated the importance of deep learning based on recent advancements towards object detection and recognition system accuracy. Cameras with different resolutions were used to gather real-time images of diseased tomato leaves. The authors used Single Shot Multibox (SSD), a region-based fully convolutional network, and a faster region-based convolutional neural network to complete their work. Single-shot multiboxes developed by the combination of two different region-based CNNs are discussed.\u003c/p\u003e"},{"header":"4. MATERIALS AND METHODS","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n \u003ch2\u003e4.1. IMAGE ACQUISITION\u003c/h2\u003e\n \u003cp\u003eIn the proposed research work, an expert-annotated benchmark dataset is considered. Data were directly captured from the field with the aid of high-end electronic devices: a Nikon Coolpix, a Sony DSLR, and an Android phone. Images captured by high-resolution cameras in a variety of dimensions are transformed to a uniform size, typically 300 by 300 pixels. Since the gathered images have different sizes, it is vital to make them all the same size. A collection of 1800 distinct photos of diseased and healthy black pepper leaves were considered for testing. These pictures were taken at random for both healthy and unhealthy leaves. For each disease type, almost 300 images were collected. Data collected from various region in Karnataka, India is depicted in Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e. Figure \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e. (a)(b)(c)(d)(e)(f)(g)(h)(i) shows field scouting in the black pepper field with an expert.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003e4.2. IMAGE PREPROCESSING\u003c/h2\u003e\n \u003cp\u003eThe visual spectrum is sensitive to the cameras used for regular photographs. There are problems with recording live photographs, including the absence of a distinct and detailed border, an ambiguous background, symptoms of various disorders, variations in the lighting conditions, an unexpected brightness, a dull or cluttered background, and others. These elements collectively have a significant impact on image analysis. Real photos need preprocessing in order to fully analyze them.\u003c/p\u003e\n \u003cp\u003eA ground truth image that is the region of interest is obtained using the grab cut algorithm. The initial step to performing the above-stated task was to mark the region of interest. To achieve this, we have used online image processing, the GIMP tool. Where we use a bounding box to mark the image. Everything outside ROI (bounding box) is considered as background and masked as zero, which is black. Next, these images are used matlab R2023a for further preprocessing. The code is written in python, importing suitable libraries. The Gaussian mixture model (GMM) is used to filter the background and foreground. GMM learns the pixel to separate background and foreground, taking into account the color pixels and the area outside the rectangular block to label the image to provide the ground truth image, Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. Later, the division is made by implementing grab-cut algorithm, where the clustering method is used by clustering all the forefront images to one node known as the source node and all the background pixels to another node known as the sink node [\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e]. These pixels are hard-labelled based on their closeness to the source node as foreground and their sink node as background. The weight among the pixels is defined by the edges defined by the rectangle block. This decides the probability of grouping pixels in each terminal. The min-cut (minimum-cut) technique is used to further partition the image graph into distinct portions using the minimum cost function. Considering the probability of the pixels being next to the source node or sink node, the image is segmented, and finally, labelled annotated image data is obtained. Sample data is shown in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003e4.3. IMAGE AUGMENTATION\u003c/h2\u003e\n \u003cp\u003eImage augmentation is a method where the images are modified into a new form of the similar image. This is done to expand the training data set. Augmentation is used to represent a given image in various versions. Such that the images are flipped, rotated, and tilted to add extra records to the given database [\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e]. To artificially increase the training dataset, the given data is vertically rotated on the x-axis, modified at a 90-degree angle, and rescaled as elucidated in some sample in Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e(a), (b), (c). This is achieved using python\u0026rsquo;s built-in functions while training a deep neural network.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e4.4. CONVOLUTIONAL NEURAL NETWORKS\u003c/h2\u003e\n \u003cp\u003eFeature extraction is the main reason for using machine learning for data organization. Convolutional Neural Network (CNN) is a type of machine learning algorithm that has been widely used for image categorization in recent years. These collection of algorithms in neural networks are termed as deep learning algorithms. These algorithms are the extensions of neural network processes. The convolutional neural network is utilized to automatically train the diseases [\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e]. Such systems are real-time deployable. Convolutional layers in CNN are followed by a pooling layer, an intermediate layer and further layers. There is an activation function in the pooling layer and between each set of the convolutional layer, Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e. The input image was processed through a filter by the convolutional layer primarily, which modifies the pixel intensities [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e]. The activation function helps in determining neurons\u0026rsquo; states, which also sends a signal to the subsequent linked neuron in the higher layers. The convolution responses are shrunk down to a smaller dimension using the pooling layer. This layer can be used with a variety of pooling techniques, including maximum, minimum, average, and other types [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e] [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e]. We have implemented a CNN to classify the black pepper plant leaf images based on disease categories and implemented max pooling (maximum) for multi class classification.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eInception V3\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eThe recommended Inception V3 neural network architecture, is 22 layers deep with 27 pooling layers encompassed, depicting the stepwise execution details. It consists of a total of nine linearly stacked inception modules. Terminals of the inception module are connected to the global average pooling layer. The deep neural network is designed at the succeeding layer by replacing the last few layers, including the final layer [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e]. For these layers, the learning rate is set high, to incorporate a faster learning rate into the newly formed (higher-level) layers. Inception V3 is employed to classify diseases in black pepper leaves effectively.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eGoogle Net\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eGoogle net was designed and proposed by google in collaboration with numerous universities. Google net consists of many different kind of 1 X 1 convolution, average pooling to create deeper architecture. The convolutional network is used to reduce the number of parameters related to weight and bias so that small network can be build up. [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e] The global average pooling used at the end\u0026rdquo; of this network, averages a feature map of 7 X 7 to 1X1. Thereby decreasing trainable parameters to 0 while improving accuracy of high class. The accuracy of google net is much more improved than its predecessors.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eSqueezeNet\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eThe squeezenet architecture is also designed to reduce the parameter of the original data matrix. It reduces the dimensions by using 1X1 convolution by the using the design strategy that normally squeezes the parameter and the design is termed as fire modules. Squeeze Net architecture are based on embedded system which are highly resource inhibited operations. The main filter used in this system is of size 1X1 and not 3 X3 that is used in most of the deep learning models. The fire model consists of squeeze convolutional layer which has filter of size 1 X 1 this is further expanded and feed into other filters that are expandable s this would expand to mixture of 1 X 1and 3X3 filters. Squeeze net model reduces the input module to smaller networks [\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e]. This small network helps in reducing the size and increasing compatibility. Squeeze net requires comparatively lesser processing time, reducing CPU inferences.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eResnet-18\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eResnet stands for residual network. This network architecture is formed on the basis of VGG-16 network. But the layers in resnet-18 are short circuited. Figure \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003ea shows the short circuit connection between the CNN layers. The concept followed here is to skip in between layers through short circuit connections. Hence this is some time called skip connections [\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]. Which are responsible for further making residual blocks. Importance is given in mapping the residual block rather than the underlying layers themselves.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e4.5. TRANSFER LEARNING\u003c/h2\u003e\n \u003cp\u003eTransfer learning is an approach that consists of domains and tasks. The domain comprises of two components. A feature space \u003cstrong\u003eH\u003c/strong\u003e is defined in the domain D with marginal probability distribution P(H), where H= {h\u003csub\u003e1....,\u003c/sub\u003e h\u003csub\u003en},\u003c/sub\u003e h\u003csub\u003ei\u003c/sub\u003e \u003cstrong\u003e\u0026isin;\u003c/strong\u003e \u003cstrong\u003eH\u003c/strong\u003e. A task in given specific domain D, specified as D = {\u003cstrong\u003eH\u003c/strong\u003e, P(F)}, consists of two components: a labelled space G and an objective predictive function \u003cstrong\u003ef\u003c/strong\u003e: \u003cstrong\u003eH\u0026loz;\u003c/strong\u003e G. Now for a new instance h, the prediction of corresponding label \u003cstrong\u003ef\u003c/strong\u003e(k) is thru function \u003cstrong\u003ef\u003c/strong\u003e. That is task T = {G, \u003cstrong\u003ef\u003c/strong\u003e(k)}, is discovered from the training data consisting of pairs {h\u003csub\u003ei,\u003c/sub\u003e g\u003csub\u003ei\u003c/sub\u003e}, where h\u003csub\u003ei\u003c/sub\u003e \u003cstrong\u003e\u0026isin;\u003c/strong\u003e \u003cstrong\u003eH\u003c/strong\u003e and g\u003csub\u003ei\u003c/sub\u003e \u003cstrong\u003e\u0026isin;\u003c/strong\u003e G\u003c/p\u003e\n \u003cp\u003eT= {G, P(G/H)} = {G, \u003cstrong\u003ef\u003c/strong\u003e(k)}\u003c/p\u003e\n \u003cp\u003eWhen there is a known source domain D\u003csub\u003eS\u003c/sub\u003e and task to learn \u003cstrong\u003eT\u003c/strong\u003e\u003csub\u003e\u003cstrong\u003eS\u003c/strong\u003e\u003c/sub\u003e, a target D\u003csub\u003eT\u003c/sub\u003e and learning task T\u003csub\u003eL\u003c/sub\u003e transfer learning accelerate the acquisition of anticipated target function \u003cstrong\u003ef\u003c/strong\u003e\u003csub\u003e\u003cstrong\u003eT\u003c/strong\u003e\u003c/sub\u003e\u003cstrong\u003e(.)\u003c/strong\u003e in D\u003csub\u003eT\u003c/sub\u003e using the knowledge in D\u003csub\u003eS\u003c/sub\u003e and \u003cstrong\u003eT\u003c/strong\u003e\u003csub\u003e\u003cstrong\u003eS .\u003c/strong\u003e\u003c/sub\u003eThe loss function generally represents the misclassification in predicted output to the input provided [\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e]. In transfer learning for classification-based optimization, the loss function is estimated as cross entropy. In machine learning cross entropy, is an approach used while the algorithms are developed to predict from the model. This is defined as given equation:\u003c/p\u003e\n \u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\text{H}( \\text{p},\\text{q})={\\sum }_{\\text{x}\\in \\text{X} }^{}p\\left(x\\right) Log q\\left(x\\right)$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eHere in the given equation x signifies the total number of values and p(x) categorizes the probability distribution in the real world and q(x) characterizes the probability of projected value in predicted environment. Overall, defining the loss function for predicted class [\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]. Transfer learning implementation details is shown in Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e, in the form of block diagram with the flow of action.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003c/div\u003e"},{"header":"5. EXPERIMENTAL SETUP","content":"\u003cp\u003eIn this study, we introduce an ensembled state-of-the art deep learning method for predicting black pepper leaf diseases. The process is carried out in two phase, first phase is data preparation and second phase involve implementation of deep neural networks for diseases classification. Images are collected from the field and annotated. The annotated/labelled images are used for the experiment [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e]. The dataset comprises of 1500 black pepper leaf images. Out of these, the best set of images are chosen for each disease category, i.e., for anthracnose, early stage phytophthora, phytophthora, slow wilt, yellowing and healthy leaf images. Hence, a total 600 images are labelled and saved [\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]. The leaf images are chosen such that the diseases are clearly evident with high visibility on the leaf surface [\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e]. The insinuation of diseases in black pepper is that they are specific to the given region, and symptoms vary from one region to another. Hence, care is taken to correctly identify and diagnose the leaf diseases. Figure \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e depicts data acquisition and annotation flow.\u003c/p\u003e\n\u003cp\u003eImage segmentation is done using grab- cut method. Grab cut is a very user-friendly and powerful algorithm for extracting the foreground objects from a marked region of interest. Image augmentation is necessary for the deep learning models to run effectively during the training phase [\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e]. Image augmentation relates to image data scaling and rotation. More information in the form of data variation enhances the performance of deep neural networks.\u003c/p\u003e\n\u003cp\u003eNext, there was a need to import big data set to train deep neural networks so that these trained networks could be used for comparing and processing our developed data set. Hence, for training the CNN model, we have casted-off the most widely used image dataset, ImageNet. This repository has 10\u0026nbsp;billion unique images in its library [\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e] [\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e]. This is beneficial for training and boosting the learning rate of deep neural network at the initial layer network. This data is loaded at the initial neural network layer in order to pretrain the neurons. To prevent misinterpretation, we gathered and considered the images from a single input source (one device) during the processes of segmentation and augmentation [\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e]. Images are resized to 256 X 256, the size expected by the input layer of each deep learning model. Large data sets are necessary for the deep learning performances; hence data augmentation is accoutered [\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e] [\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e]. These are considered the new databases for transfer learning implementation. Figure \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e (a) represents Resnet residual block and Fig. \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e(b) represents CNN implementation. The flow of proposed conceptual methodology using transfer learning is illustrates in Fig. \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e.\u003c/p\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003e5.1. ARCHITECTURE FOR PROPOSED DEEP LEARNING IMPLEMENTATION\u003c/h2\u003e\n \u003cp\u003eContemporary state-of-art deep neural network are the best models to utilize for any given task. Based on its accuracy, speed or any other parameters of importance, a DNN (deep neural network) might be classified as SOTA (State-of-the-art). The majority of computer vision fields, though, involve a trade-off between these measurements. In other words, even if a DNN is extremely quick, its accuracy may not be up to mark. Sometimes we can create a model with acceptable performance metrics, but it doesn\u0026rsquo;t have the necessary latency or throughput for a variety of applications, such as picture categorization and object detection. In CNN, features are extracted from training data using the convolutional layer.\u003c/p\u003e\n \u003cp\u003eIn this experiment, after meticulously training and working on new benchmark data set, four different deep neural networks: Inception V3, GoogleNet, SqueezeNet, and ResNet-18 are trained as state-of-the-art model for the black pepper leaf disease classification, using Matlab R2023a software. First, the data is trained using Inception V3 individually with and without transfer learning, and the results are noted. Next an ensemble of state-of-the-art deep learning models that is GoogleNet, SqueezeNet and ResNet-18 are used with the optimizers;\u0026rdquo; adam\u0026rdquo;,\u0026rdquo; sgdm\u0026rdquo; and \u0026ldquo;rmsprop\u0026rdquo;. The hyperparameters used are data partitioning, loss functioning-batch size, initial learning rate schedule, learning rate, learn rate factor, number of epochs, validation frequency and bias learn rate factor.\u003c/p\u003e\n \u003cp\u003eThe activation function used is ReLU, which is the most prevalent function in a deep neural network. Our classification problem is based on multiclass classification, hence SoftMax is employed. The code is written in python, importing all the necessary python packages in MATLAB. Further these networks are fine-tuned and learned higher-level layers are replaced by new layer including max-pooling, for the benchmark dataset classification. We have used cross entropy for improved classification probability and threshold cut to avoid overfitting. The accuracy obtained through transfer learning technique is very high. The proposed data is classified into: healthy and a set of diseased images. The six-leaf disease classification include Anthracnose, Early stage phytophthora, Slow wilt, Yellowing, Phytophthora and. Performance of the ensembled state-of-the-art network is measured by confusion metrics.\u003c/p\u003e\n \u003cp\u003eNot much work is done in automating early leaf disease detection in black pepper leaves. Our approach is one of a kind for black pepper leaf disease detection. The implementation of deep learning algorithm has increased the performance of the classification model both in terms of accuracy and processing time using transfer learning technique. This work represents early disease classification and prediction with state-of-the-art deep learning models. This work represents implementation of high-end algorithms towards the automation of agricultural unit for the benefit of farmers and people working in fields. This is also an attempt to contribute to society, as the results obtained can be used in precision agriculture in the future. The architecture of the proposed work is depicted in Fig. \u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" style=\"width: 891px;\" width=\"891\" height=\"699\"\u003e\u003cbr\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003ch2\u003e5.2.PERFORMANCE METRIX\u003c/h2\u003e\n \u003cp\u003eThe efficiency of the proposed black pepper leaf disease detection is measured by the confusion matrix. Which gives performance measures based on true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). The confusion matrix is represented as in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eConfusion Matrix\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eConfusion Matrix\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003ePredicted Class\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eActual Class\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTrue Positive (TP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFalse Positive (FP)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFalse Negative (FN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTrue Negative (TN)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003eThe performance that are computed through confusion matrix are specificity, sensitivity, precision, accuracy and F1-score.\u003c/p\u003e\n \u003cp\u003eSpecificity:\u003c/p\u003e\n \u003cp\u003eIt is the performance measure, which represents the ratio of true negative to that of false positive and true negative.\u003c/p\u003e\n \u003cp\u003eSpecificity = {TN}/{FP\u0026thinsp;+\u0026thinsp;TN}\u003c/p\u003e\n \u003cp\u003eSensitivity:\u003c/p\u003e\n \u003cp\u003eIt is the performance measure, which represents the ratio of true positive to that of true positive and false negative. It is also equivalent to recall.\u003c/p\u003e\n \u003cp\u003eSensitivity\u0026thinsp;=\u0026thinsp;Recall = {TP}/{TP\u0026thinsp;+\u0026thinsp;FN}\u003c/p\u003e\n \u003cp\u003ePrecision:\u003c/p\u003e\n \u003cp\u003ePrecision represents the ratio of true positive to that of true positive and false positive. It is also equivalent to recall.\u003c/p\u003e\n \u003cp\u003ePrecision = {TP}/{TP\u0026thinsp;+\u0026thinsp;FP}\u003c/p\u003e\n \u003cp\u003eAccuracy:\u003c/p\u003e\n \u003cp\u003eThis performance measure is defined as the ratio of sum of true positive and true negative to that of total sum of true positive, false positive, true negative and false negative.\u003c/p\u003e\n \u003cp\u003eAccuracy = {TP\u0026thinsp;+\u0026thinsp;TN}/{TP\u0026thinsp;+\u0026thinsp;TN\u0026thinsp;+\u0026thinsp;FP\u0026thinsp;+\u0026thinsp;FN}\u003c/p\u003e\n \u003cp\u003eF1-score: This performance metric is the harmonic mean of recall and precision. It is measure of the accuracy. Represented as given equation.\u003c/p\u003e\n \u003cp\u003eF1 = {2*Precision*Recall} {Precision Recall} = {2*TP} {2*TP\u0026thinsp;+\u0026thinsp;FP\u0026thinsp;+\u0026thinsp;FN}\u003c/p\u003e\n\u003c/div\u003e"},{"header":"6. RESULTS AND DISCUSSION","content":"\u003cp\u003eWe have used MATLAB R2023a for our deep neural network implementation. The deep neural network is customized accordingly to get better performance. An ensemble of deep learning models is used for six black pepper leaf disease classification. Figure \u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e represents the architecture of proposed model. Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e shows various hypermeters used in the experiment and its definition with values. A comparative analysis is done with other deep learning models given in various research paper is shown in Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e.A comparison is also shown in Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e with the proposed model accuracy obtained in the experiment. The stage wise steps are specified and explained through algorithm: Implementing ensemble state-of-art deep learning model for prediction of black pepper leaves diseases. A total 9 different model combination is made with three different optimizers \u0026ldquo;adam\u0026rdquo;,\u0026rdquo; sgdm\u0026rdquo; and \u0026ldquo;msprop.\u0026rdquo;. The images are resized at pre-processing stage as required by the deep model algorithm i.e. 224 \u0026times;224 \u0026times;3 and 227 \u0026times; 227\u0026times;3, etc. The learning rate used in the algorithm is 0.001 and 0.0001. The predicted output acquired through MATLAB tool is shown in Fig. \u003cspan class=\"InternalRef\"\u003e11\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eImage augmentation and segmentation is done during preprocessing stage. This helped in having better accuracy and F1 score from0.78 to 0.98. Ensemble deep learning approach with transfer learning technique resulted in benefiting high prediction of black pepper disease classification from 65\u0026ndash;99%. This is seen with Inception V3 implementation in both cases. Table \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e \u0026amp; Table \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e exhibit architecture layer details and analysis result. The accuracy graph obtained is depicted in Fig. \u003cspan class=\"InternalRef\"\u003e12\u003c/span\u003e(Inception V3). The performance of the state-of-the-art deep learning model is measured through confusion matrix. The resulted confusion matrix in MATLAB tool is shown in Fig. \u003cspan class=\"InternalRef\"\u003e13\u003c/span\u003e \u0026amp; Fig. \u003cspan class=\"InternalRef\"\u003e14\u003c/span\u003e. Where the comparison is shown for training as well as testing data. Figure \u003cspan class=\"InternalRef\"\u003e13\u003c/span\u003e demonstrate the accuracy graph for Inception V3 and Fig. \u003cspan class=\"InternalRef\"\u003e14\u003c/span\u003e represents for rest of the models. The performance measure acquired by the three models \u0026ldquo;GoogleNet\u0026rdquo;,\u0026rdquo; SqueezeNet\u0026rdquo; and \u0026ldquo;ResNet-18\u0026rdquo; is shown in Fig. \u003cspan class=\"InternalRef\"\u003e15\u003c/span\u003e and Fig. 16, which is the confusion matrix for training and test set. The values are computed through confusion matrix. The statistic is displayed in Table \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e. Five different disease classes are considered with one healthy class. Figure \u003cspan class=\"InternalRef\"\u003e17\u003c/span\u003e exhibits the calculated sensitivity, specificity, precision, accuracy and F1 score for the selected deep learning models. Figure \u003cspan class=\"InternalRef\"\u003e18\u003c/span\u003e depicts the percentage accuracy of the prediction made for various category of leaf class by the model, based on confusion matrix. It is also observed that performance of the deep leaning models varies with the optimizers used. GoogleNet works well with rmsprop and sgdm. SqueezeNet works better with sgdm optimizer. ResNet-18 results in high accuracy with rmsprop optimizer. The proposed modified Res-18 outperforms all other deep learning models used in the experiment. But the outcome may vary for different dataset other than ImageNet and proposed diseased Black pepper leaf dataset. The efficiency of each available high-end deep neural network algorithm depends experiment and analysis.\u003c/p\u003e\n\u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eHyperparameters used in the experiment\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eS. No\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eHyperparameters\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eValue/Parameters\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eData partition\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIt relates to the division made w.r.t training and testing dataset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eData divided as ratio of :\u003c/p\u003e\n \u003cp\u003e1.Training\u0026thinsp;=\u0026thinsp;70%\u0026amp;Testing 30%\u003c/p\u003e\n \u003cp\u003e2.Training\u0026thinsp;=\u0026thinsp;60%\u0026amp;Testing 40%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInput image size\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eThe image size required by the model being trained\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.224 \u0026times; 224 \u0026times; 3\u003c/p\u003e\n \u003cp\u003e2.227\u0026times; 227 \u0026times; 3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eloss function\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRelates to error occurred with given value to that of algorithm output\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0 to 1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emini-batch size\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNormally the batch sizes considered are power of 2. Not recommended for dataset less \u0026lt;\u0026thinsp;2000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e64,128256,512 etc.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003einitial learning rate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIt is the value when neural network starts learning.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.001/0.0001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003elearning rate schedule\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eA parameter to adjust networks learning to that of predefined value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.Time decay\u003c/p\u003e\n \u003cp\u003e2.Step decay\u003c/p\u003e\n \u003cp\u003e3.Exponential decay\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enumber of epochs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRelates to the iteratively working/training on entire batch of data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10,25,30\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003elearn rate factor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIt is non-negative scalar value. Where learning rate of specific parameter is compared to global learning rate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ebias learn rate factor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRate at which neural network learns\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003evalidation patience\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIt is the maximum tries the algorithm makes w..r.t epoch for performance improvement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003evalidation frequency\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIt is the lowest value by which the given model is to validated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eperformance metrics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMeasure of the classification performance of machine learning algorithms\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eConfusion Metrix\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOptimizers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eA function that familiarizes the neural network\u0026rsquo;s weights and learning rate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAdam, Sgdm, Rmsprop\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison result of other deep learning models\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eReference\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassification Model\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eType of Leaf\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGanatra, N. and Patel, A [\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMultiple\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e73.38%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSaleem MH [\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eXception architecture\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMultiple\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e91.86\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eShreya Ghosal [\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVGG-16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRice Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e92.46%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHassan, S k Mahmudul [\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInceptionResNetV2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlant Village Dataset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.02%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eArvind Krishnaswamy R [\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVGG-16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSelected Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e90.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSantosh Kumar Up [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePaddy leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e97.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSaleem, Muhammad H [\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDeep Neural Network\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTomato Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e95.76%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGeetharamani, G [\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDeep CNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlant Leaf Dataset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e96.46%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003ctable id=\"Tab4\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of state of art deep learning approach for black pepper leaves\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDeep Neural Network\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassification Model\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eLeaf Type\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception V3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception V3\u003c/p\u003e\n \u003cp\u003e(without transfer learning)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlack Pepper Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e62.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception V3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception V3\u003c/p\u003e\n \u003cp\u003e(with transfer learning)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlack Pepper Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e94.6%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGoogleNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGoogleNet\u003c/p\u003e\n \u003cp\u003e(with transfer learning)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlack Pepper Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e99.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSqueezeNet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSqueezeNet\u003c/p\u003e\n \u003cp\u003e(with transfer learning)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlack Pepper Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e99.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProposed Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eResNet-18\u003c/p\u003e\n \u003cp\u003e(with transfer learning)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBlack Pepper Leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e99.67%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003ctable id=\"Tab5\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eInception V3 detailed parameter and architecture layer\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eType\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePatch size/stride\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eOutput size\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003edepth\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e#1 \u0026times; 1\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e#3 \u0026times;3 reduce\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e#3 \u0026times; 3\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e# 5 \u0026times; 5\u003c/p\u003e\n \u003cp\u003ereduce\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e#5 \u0026times; 5\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePool proj\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eparams\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eops\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003econvolution\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7\u0026times; 7/2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e112\u0026times; 112\u0026times; 64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.7k\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003emax pool\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3\u0026times; 3/2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e56\u0026times; 56\u0026times; 64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003econvolution\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3\u0026times; 3/1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e56\u0026times; 56\u0026times; 192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e112K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e360M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003emax pool\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3\u0026times; 3/2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28\u0026times; 28\u0026times; 192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(3a)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28\u0026times; 28\u0026times; 256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e159K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e128M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eInception(3b)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28\u0026times; 28\u0026times; 480\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e380K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e304M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003emax pool\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3\u0026times; 3/2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14\u0026times; 14\u0026times; 480\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(4a)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14\u0026times; 14\u0026times; 512\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e208\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e364K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e73M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(4b)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14\u0026times; 14\u0026times; 512\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e160\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e112\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e437K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e88M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(4c)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14\u0026times; 14\u0026times; 512\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e463K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(4d)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14\u0026times; 14\u0026times; 528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e112\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e144\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e288\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e580K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e119M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(4e)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14\u0026times; 14\u0026times; 832\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e160\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e320\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e840K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e170M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003emax pool\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3\u0026times; 3/2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7\u0026times; 7\u0026times; 832\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(5a)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7\u0026times; 7\u0026times; 832\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e160\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e320\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1072K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e54M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003einception(5b)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7\u0026times; 7\u0026times; 1024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e384\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e384\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1388K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e71M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eavg pool\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7\u0026times; 7/1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1\u0026times; 1\u0026times; 1024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003edropout(40%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1\u0026times; 1\u0026times; 1024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003elinear\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1\u0026times; 1\u0026times; 1000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1000K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003esoftmax\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1\u0026times; 1\u0026times; 1000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003ctable id=\"Tab6\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eInception V3 Analysis Result\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eName\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eActivations\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eType\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eLearnables\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b_1\u0026times;1\u003c/p\u003e\n \u003cp\u003e384 1\u0026times;1\u0026times; 832 convolutions with stride [1 1] and padding [0 0 0 0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 384\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eConvolution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWeights 1 \u0026times;1\u0026times; 832 \u0026times;384\u003c/p\u003e\n \u003cp\u003eBias 1 \u0026times;1 \u0026times;384\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b-relu_1\u0026times;1\u003c/p\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 384\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b-5\u0026times;5_reduce\u003c/p\u003e\n \u003cp\u003e48 1\u0026times;1\u0026times;832 convolutions with stride [1 1] and padding [0 0 0 0]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003econvolution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWeights 1 \u0026times;1\u0026times; 832 \u0026times;48\u003c/p\u003e\n \u003cp\u003eBias 1 \u0026times;1 \u0026times;48\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b-relu_5\u0026times;5_reduce\u003c/p\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e--\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b-relu_5\u0026times;5_reduce\u003c/p\u003e\n \u003cp\u003e128 5\u0026times;5\u0026times; 48 convolutions with stride [1 1] and padding [2 2 2 2]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eConvolution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWeights 5 \u0026times;5\u0026times; 48 \u0026times; 128\u003c/p\u003e\n \u003cp\u003eBias 1 \u0026times;1 \u0026times; 128\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b-relu_5\u0026times;5\u003c/p\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e--\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b-relu_3\u0026times;3_reduce\u003c/p\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 192\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eReLU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e--\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInception_5b-3\u0026times;3\u003c/p\u003e\n \u003cp\u003e384 3\u0026times;3\u0026times; 192 convolutions with stride [1 1] and padding [1 1 1 1]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 \u0026times;7 \u0026times; 384\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eConvolution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWeights 3 \u0026times;3\u0026times; 192 \u0026times; 384\u003c/p\u003e\n \u003cp\u003eBias 1 \u0026times;1 \u0026times; 384\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable id=\"Tab7\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eStatistic Measure for state -of -the-art deep learning model\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMeasure\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003esensitivity\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003especificity\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eprecision\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eF1 Score\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDerivations\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eTPR\u0026thinsp;=\u0026thinsp;TP / (TP\u0026thinsp;+\u0026thinsp;FN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eSPC\u0026thinsp;=\u0026thinsp;TN / (FP\u0026thinsp;+\u0026thinsp;TN)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003ePPV\u0026thinsp;=\u0026thinsp;TP / (TP\u0026thinsp;+\u0026thinsp;FP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eACC = (TP\u0026thinsp;+\u0026thinsp;TN) / (P\u0026thinsp;+\u0026thinsp;N)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eF1\u0026thinsp;=\u0026thinsp;2TP / (2TP\u0026thinsp;+\u0026thinsp;FP\u0026thinsp;+\u0026thinsp;FN)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePredicted Classification\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAnthracnose (ANT)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9565\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9894\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9429\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9843\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9496\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEarlyStagePhytophthora (ESP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9753\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9959\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9875\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHealthy (HLT)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9277\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9862\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9625\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePhytophthora (PHY)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9178\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9841\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9178\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9734\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9178\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSlow Wilt (SW)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8649\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9812\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9014\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9620\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8828\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYellowing(Y)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9552\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9658\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8312\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.9642\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.8889\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"7. CONCLUSION AND FUTURE WORK","content":" \u003cp\u003eThis research focuses on the early stage diagnosis of black pepper plant leaf diseases with the comprehension of computer vision. This is the first attempt of its kind to detect black pepper leaf diseases using a deep learning approach. The proposed technique helps in predicting early symptoms of various black pepper leaf diseases. For this study, five types of black pepper leaf diseases were considered: anthracnose, early stage phytophthora, quick wilt, yellowing, and slow wilt, as well as healthy leaves. The state-of-the-art deep neural networks are implemented successfully on the newly created leaf dataset with an accuracy of 99.7%. The accuracy alters with respect to optimizers used. Deep learning models work well when data is available in millions. ResNet-18 provided high accuracy 99.7% when compared with other models.\u003c/p\u003e \u003cp\u003eThis work proposes a solution towards agricultural intelligence for predicting diseases remotely with the application of high potential deep neural network in precision agriculture. The technique can be further developed as an app for smart phones. An automated leaves prediction system can also be used by non-botanical experts to quickly identify early plants diseases quite effortlessly. Deep learning can aid the task of remotely sensing fields where human intervention is vulnerable.\u003c/p\u003e \u003cp\u003eIn the majority of circumstances, the traditional approach to diagnosis disease is similar. This may lead to misinterpretation about the diseases if it is unknown. Which may further tip to uneven application of pesticides in the fields, resulting in crop loss. Though diseases can be predicted based on climate conditions, their symptoms cannot be judged. With the help of the proposed method early disease predictions is possible. This will aid in reduction of unwanted pesticides distribution. A web-based or mobile-based computer system for the automatic classification of medicinal plants such as black pepper will help the local population to improve their knowledge of medicinal plants.\u003c/p\u003e \u003cp\u003eIn the future, region-growing segmentation techniques can be used interactively for profound disease spot detection in leaves. Bountiful advanced deep learning technologies could be explored, such as VGG-101, VGG-S, U-Net, V-Net, Yolo, and Fuzzy Logic, for leaf disease detection, classification and prediction. Zone-based or global feature-based feature extraction techniques could also be explored. Synthetically developed image data can also be used in future, provided the required input data is in few hundreds.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCompeting interest:\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors Contribution Statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnita S Kini wrote the main manuscript text, including conceptualization, Data collection and Formal Analysis, Programming and Validation.Supervision for writing is done by Prema K V \u0026amp; Smitha N Pai.All authors have reviewed the manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical \u0026amp; Informed consent for the data used\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data used in the experiment were all obtained by the first author through field scouting.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability and access\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data used to support the given experiment are available on request through the first author\u0026rsquo;s Email.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBarbedo, J.G.A., Plant disease identification from individual lesions and spots using deep learning. Biosystems Engineering, 180, 2019, pp.96-107.\u003c/li\u003e\n\u003cli\u003eToo, E.C., Yujian, L., Njuki, S. and Yingchun, L., A comparative study of fine-tuning deep learning models for plant disease identification. Computers and Electronics in Agriculture, 161, 2019, pp.272-279.\u003c/li\u003e\n\u003cli\u003eBarbedo, J.G.A. Impact of dataset size and variety on the effectiveness of deep learning and transfer learning for plant disease classification. Computers and electronics in agriculture, 153, 2018. , pp.46-53.\u003c/li\u003e\n\u003cli\u003eHossain, M.A. and Sajib, M.S.A. Classification of image using convolutional neural network (CNN). Global Journal of Computer Science and Technology. 2019.\u003c/li\u003e\n\u003cli\u003eThangaraj, Rajasekaran, S. Anandamurugan, and Vishnu Kumar Kaliappan. \u0026quot;Automated tomato leaf disease classification using transfer learning-based deep convolution neural network.\u0026quot; Journal of Plant Diseases and Protection 128, no. 1 (2021): 73-86.\u003c/li\u003e\n\u003cli\u003eGhosal, Shreya, and Kamal Sarkar. \u0026quot;Rice leaf diseases classification using CNN with transfer learning.\u0026quot; In 2020 IEEE Calcutta Conference (CALCON), pp. 230-236. IEEE, 2020.\u003c/li\u003e\n\u003cli\u003eGanatra, N. and Patel, A., 2020. A multiclass plant leaf disease detection using image processing and machine learning techniques. Int. J. Emerg. Technol, 11(2), pp.1082-1086.\u003c/li\u003e\n\u003cli\u003eSaleem MH, Potgieter J, Arif KM. Plant disease classification: A comparative evaluation of convolutional neural networks and deep learning optimizers. Plants. 2020 Oct;9(10):1319.\u003c/li\u003e\n\u003cli\u003eSagar, Abhinav, and Dheeba Jacob. \u0026quot;On using transfer learning for plant disease detection.\u0026quot; bioRxiv (2021): 2020-05. \u003c/li\u003e\n\u003cli\u003eHasan, Reem Ibrahim, Suhaila Mohd Yusuf, and Laith Alzubaidi. \u0026quot;Review of the state of the art of deep learning for plant diseases: a broad analysis and discussion.\u0026quot; Plants 9, no. 10 (2020): 1302.\u003c/li\u003e\n\u003cli\u003eHassan, Sk Mahmudul, Arnab Kumar Maji, Michał Jasiński, Zbigniew Leonowicz, and Elżbieta Jasińska. \u0026quot;Identification of plant-leaf diseases using CNN and transfer-learning approach.\u0026quot; Electronics 10, no. 12 (2021): 1388.\u003c/li\u003e\n\u003cli\u003eDevaraj, Abirami, Karunya Rathan, Sarvepalli Jaahnavi, and K. Indira. \u0026quot;Identification of plant disease using image processing technique.\u0026quot; In 2019 International Conference on Communication and Signal Processing (ICCSP), pp. 0749-0753. IEEE, 2019.\u003c/li\u003e\n\u003cli\u003eAravind, Krishnaswamy Rangarajan, and Purushothaman Raja. \u0026quot;Automated disease classification in (Selected) agricultural crops using transfer learning.\u0026quot; Automatika 61, no. 2 (2020).\u003c/li\u003e\n\u003cli\u003eA. S. Kini, P. K. V and S. N. Pai, \u0026quot;Leaves Classification Through Image Segmentation and Convolutional Neural Network,\u0026quot; 2021 3rd\u003cem\u003e \u003c/em\u003eInternational Conference on Advances in Computing, Communication Control and Networking (ICAC3N\u003cem\u003e)\u003c/em\u003e, Greater Noida, India, 2021, pp. 631-634, doi: 10.1109/ICAC3N53548.2021.9725744.\u003c/li\u003e\n\u003cli\u003eA. S. Kini, K. Prema and S. N. Pai, \u0026quot;Intelligent classification model for Indian chickpea,\u0026quot; 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 2021, pp. 1847-1852, doi: 10.1109/ICICCS51141.2021.9432129.\u003c/li\u003e\n\u003cli\u003eUpadhyay, Santosh Kumar, and Avadhesh Kumar. \u0026quot;Early-Stage Brown Spot Disease Recognition in Paddy Using Image Processing and Deep Learning Techniques.\u0026quot; Traitement du Signal 38, no. 6 (2021): 1755-1766.\u003c/li\u003e\n\u003cli\u003eHasan, Mosin, Bhavesh Tanawala, and Krina J. Patel. \u0026quot;Deep learning precision farming: Tomato leaf disease detection by transfer learning.\u0026quot; In Proceedings of 2nd international conference on advanced computing and software engineering (ICACSE). 2019.\u003c/li\u003e\n\u003cli\u003eLi, Lili, Shujuan Zhang, and Bin Wang. \u0026quot;Plant disease detection and classification by deep learning\u0026mdash;a review.\u0026quot; IEEE Access 9 (2021): 56683-56698.\u003c/li\u003e\n\u003cli\u003eArsenovic, Marko, Mirjana Karanovic, Srdjan Sladojevic, Andras Anderla, and Darko Stefanovic. \u0026quot;Solving current limitations of deep learning-based approaches for plant disease detection.\u0026quot; Symmetry 11, no. 7 (2019): 939. \u003c/li\u003e\n\u003cli\u003eSaleem, Muhammad Hammad, Johan Potgieter, and Khalid Mahmood Arif. \u0026quot;Plant disease detection and classification by deep learning.\u0026quot; Plants 8, no. 11 (2019): 468.\u003c/li\u003e\n\u003cli\u003eSaleem, Muhammad Hammad, Johan Potgieter, and Khalid Mahmood Arif. \u0026quot;Plant disease classification: A comparative evaluation of convolutional neural networks and deep learning optimizers.\u0026quot; Plants 9, no. 10 (2020): 1319. \u003c/li\u003e\n\u003cli\u003ePaymode, Ananda S., and Vandana B. Malode. \u0026quot;Transfer learning for multi-crop leaf disease image classification using convolutional neural networks VGG.\u0026quot; Artificial Intelligence in Agriculture (2022). \u003c/li\u003e\n\u003cli\u003eKaya, Aydin, Ali Seydi Keceli, Cagatay Catal, Hamdi Yalin Yalic, Huseyin Temucin, and Bedir Tekinerdogan. \u0026quot;Analysis of transfer learning for deep neural network-based plant classification models.\u0026quot; Computers and electronics in agriculture 158 (2019): 20-29.\u003c/li\u003e\n\u003cli\u003eGeetharamani, G., and Arun Pandian. \u0026quot;Identification of plant leaf diseases using a nine-layer deep convolutional neural network.\u0026quot; Computers \u0026amp; Electrical Engineering 76 (2019): 323-338.\u003c/li\u003e\n\u003cli\u003eBarth, Ruud, Joris IJsselmuiden, Jochen Hemming, and Eldert J. Van Henten. \u0026quot;Synthetic bootstrapping of convolutional neural networks for semantic plant part segmentation.\u0026quot; Computers and Electronics in Agriculture 161 (2019): 291-304.\u003c/li\u003e\n\u003cli\u003eKini, A.S., Reddy, P.K. and Pai, S.N., 2023. Techniques of deep learning and image processing in plant leaf disease detection: a review. International Journal of Electrical and Computer Engineering, 13(3), pp.3029-3040\u003cem\u003e.\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eRajesh, B., M. Vishnu Sai Vardhan, and L. Sujihelen. \u0026quot;Leaf Disease Detection and Classification by Decision Tree.\u0026quot; In 2020 4th International Conference on Trends in Electronics and Informatics (ICOEI) (48184), pp. 705-708. IEEE, 2020.\u003c/li\u003e\n\u003cli\u003eCoulibaly, Solemane, Bernard Kamsu-Foguem, Dantouma Kamissoko, and Daouda Traore. \u0026quot;Deep neural networks with transfer learning in millet crop images.\u0026quot; Computers in Industry 108 (2019): 115-120.\u003c/li\u003e\n\u003cli\u003eVilasini, M. \u0026quot;The CNN Approaches for Classification of Indian Leaf Species Using Smartphones.\u0026quot; Computers, Materials \u0026amp; Continua 62, no. 3 (2020): 1445-1472.\u003c/li\u003e\n\u003cli\u003eLiu, Jun, and Xuewei Wang. \u0026quot;Plant diseases and pests detection based on deep learning: a review.\u0026quot; Plant Methods 17, no.1 (2021): 1-18.\u003c/li\u003e\n\u003cli\u003eFuentes, Alvaro, Sook Yoon, Sang Cheol Kim, and Dong Sun Park. \u0026quot;A robust deep-learning-based detector for real-time tomato plant diseases and pest\u0026rsquo;s recognition.\u0026quot; Sensors 17, no. 9 (2017): 2022.\u003c/li\u003e\n\u003cli\u003eYang, Kunlong, Weizhen Zhong, and Fengguo Li. \u0026quot;Leaf segmentation and classification with a complicated background using deep learning.\u0026quot; Agronomy 10, no. 11 (2020): 1721.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Black pepper, Convolutional neural network, Deep learning, Early-stage Leaf diseases, Image segmentation, Transfer Learning","lastPublishedDoi":"10.21203/rs.3.rs-3272019/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3272019/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eBlack pepper is a medicinal plant that is extensively used in Ayurvedic medicine because of its therapeutic properties. Leaf diseases can be diagnosed at an early stage with the aid of a smart computer vision system and timely disease prevention can be targeted. The proposed work represents an intelligent transfer learning technique through state-of-the-art deep learning application to predict the presence of prominent diseases in black pepper leaves. The ImageNet dataset available online is used for training deep neural network, initially. Later, this trained network is utilized for the prediction of the developed black pepper leaf image dataset. The developed data set consist of real time leaf images, which are candidly taken from the fields and annotated under supervision of an expert. The leaf diseases considered including healthy leaves are anthracnose, slow wilt, early stage phytophthora, phytophthora and yellowing. The accuracy obtained with 0.001 learning rate ranges from 99.1\u0026ndash;99.5% for the Inception V3, GoogleNet, SqueezeNet and Resnet18 models. This work represents improvement in agriculture and a cutting edge deep neural network method for early stage leaf disease identification and prediction. This is an approach using a deep learning to predict black pepper leaf disease.\u003c/p\u003e","manuscriptTitle":"State of the Art Deep Learning Implementation for Multiclass Classification of Black Pepper Leaf Diseases","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-08-25 19:05:31","doi":"10.21203/rs.3.rs-3272019/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2023-11-16T14:09:00+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-11-07T14:55:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"885a1a21-9332-4e7a-a829-96f562d08fa6","date":"2023-10-18T07:58:00+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-09-24T05:49:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"4cc85946-9810-4b5d-b803-601cd57fe53d_SNPRID","date":"2023-09-15T12:55:03+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2023-09-15T12:49:24+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2023-09-15T12:40:47+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2023-08-23T02:15:15+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2023-08-23T02:08:18+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2023-08-17T11:21:18+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"987ab309-b971-45eb-a138-a6c4a5105b5c","owner":[],"postedDate":"August 25th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":24221030,"name":"Physical sciences/Engineering/Electrical and electronic engineering"},{"id":24221031,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"},{"id":24221032,"name":"Biological sciences/Computational biology and bioinformatics/Computational platforms and environments"},{"id":24221033,"name":"Biological sciences/Computational biology and bioinformatics/Data acquisition"},{"id":24221034,"name":"Biological sciences/Computational biology and bioinformatics/Data integration"},{"id":24221035,"name":"Biological sciences/Computational biology and bioinformatics/Data processing"},{"id":24221036,"name":"Biological sciences/Computational biology and bioinformatics/Image processing"},{"id":24221037,"name":"Biological sciences/Computational biology and bioinformatics/Machine learning"},{"id":24221038,"name":"Earth and environmental sciences/Environmental sciences"},{"id":24221039,"name":"Physical sciences/Energy science and technology"},{"id":24221040,"name":"Physical sciences/Engineering"},{"id":24221041,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2024-01-22T15:06:07+00:00","versionOfRecord":{"articleIdentity":"rs-3272019","link":"https://doi.org/10.1038/s41598-024-51884-0","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2024-01-16 15:01:09","publishedOnDateReadable":"January 16th, 2024"},"versionCreatedAt":"2023-08-25 19:05:31","video":"","vorDoi":"10.1038/s41598-024-51884-0","vorDoiUrl":"https://doi.org/10.1038/s41598-024-51884-0","workflowStages":[]},"version":"v1","identity":"rs-3272019","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3272019","identity":"rs-3272019","version":["v1"]},"buildId":"7rjqhiLT3MXkJMwkYKINL","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}