Comparative Analysis of Deep Learning Models for Plant Disease Detection

Comparative Analysis of Deep Learning Models for Plant Disease Detection | 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 Research Article Comparative Analysis of Deep Learning Models for Plant Disease Detection Bhageerathi T, Anagha M, Pushpa T S This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5348075/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The detection of plant diseases through deep learning models repre- sents a significant advancement in agricultural management. This study pro- vides a comprehensive accuracy comparison of four prominent deep learning models—Convolutional Neural Networks (CNN), AlexNet, DenseNet, and VGG16—for identifying plant diseases from leaf images. Leveraging the PlantVillage dataset, which includes over 11,254 images of healthy and dis- eased leaves, the research investigates the strengths and limitations of each model in terms of accuracy, feature extraction, and classification performance. DenseNet's densely connected architecture and VGG16's deep layers are high- lighted for their superior ability to handle complex patterns in diseased leaves. The study demonstrates that DenseNet achieves the highest accuracy, making it a viable solution for real-time disease detection in precision agriculture. By comparing these models, the research aims to guide the selection of the most effective deep learning approach for improving plant health monitoring. Plant Disease Detection Deep Learning Convolutional Neural Networks (CNN) AlexNet DenseNet VGG16 Image Classification PlantVillage Dataset Precision Agriculture Automated Diagnosis Model Comparison Accuracy 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 1. Introduction In the rapidly evolving field of agriculture, the health and productivity of crops are paramount to ensuring food security and economic stability. Plant diseases pose a significant threat to agricultural output, leading to substantial losses each year. Traditional methods of disease detection, which rely on visual inspection by experts, are often time-consuming, subject to human error, and not scalable across large fields. As a result, there is an increasing demand for innovative solutions that can automate and enhance the accuracy of disease detection. Deep learning has emerged as a powerful tool in the domain of computer vision, enabling the development of models that can automatically detect and classify plant diseases from leaf images. By leveraging large datasets and advanced neural network architectures, these models can identify complex patterns and subtle variations in leaf texture and color that are indicative of specific diseases.This research focuses on comparing the accuracy and performance of four widely-used deep learning models—Convolutional Neural Networks (CNN), AlexNet, DenseNet, and VGG16—in the context of plant disease detection. Utilizing the PlantVillage dataset, which comprises over 11,254 images of healthy and diseased leaves, the study aims to evaluate the strengths and limitations of each model. The investigation emphasizes the ability of DenseNet's densely connected architecture and VGG16's deep layers to manage the intricate patterns present in diseased leaves. II. Literature Review The paper discusses a deep learning-based model for crop disease detection us- ing the PlantVillage dataset. It compares the performance of MobileNet and In- ceptionV3 models, highlighting that InceptionV3 achieved higher accuracy in detecting crop species and diseases. The study suggests that the methodology could be extended to smartphone applications for practical use[1]. The paper presents a crop disease detection system using the YOLO (You Only Look Once) algorithm, which excels in real-time image processing at 45 frames per second. By employing YOLO, which divides images into grid cells to pre- dict bounding boxes and class probabilities in one evaluation, the system offers faster and more accurate detection of plant diseases compared to traditional methods. The study highlights YOLO’s efficiency in processing leaf images to identify diseases quickly and effectively, aiming to improve crop management for farmers[2]. The paper presents a comparative study of various algorithms for plant leaf dis- ease detection using deep learning techniques. It evaluates different models based on their performance in accurately classifying diseases in plant leaves. The study highlights the importance of choosing the right model for effective plant disease diagnosis, emphasizing the impact of different architectures on de- tection accuracy[3]. The paper describes an automated system for detecting crop diseases using remote sensing images. The approach consists of two phases: training and mon- itoring. The training phase involves collecting images of both healthy and dis- eased crops, extracting threshold values, and using RGB layer separation. The monitoring phase uses these threshold values to compare with new images and apply techniques like Canny edge detection and histogram analysis to identify diseases. If a disease is detected, the system sends an alert to the farmer via SMS. The paper demonstrates the method's effectiveness through experimental results using MATLAB[4]. The paper introduces a lightweight CNN model for plant disease identification using Inception and Residual connections with depthwise separable convolu- tions to reduce parameters and computational complexity. The model outper- forms state-of-the-art models on three datasets, achieving high accuracy with fewer resources. This approach addresses the challenges of deploying deep learning models in resource-constrained environments[5]. Crop diseases and pests significantly impact agricultural productivity, mak- ing their detection crucial. This paper reviews deep learning techniques for de- tecting crop diseases and pests and proposes a Convolutional Neural Network (CNN) model for automatic diagnosis. The methodology involves image acqui- sition, preprocessing, feature extraction, and classification using CNN with transfer learning. This approach enhances detection accuracy, aiding precision agriculture[6]. III. Objectives Develop and Compare Models: Build and evaluate Convolutional Neural Net- works (CNN), AlexNet, DenseNet, and VGG16 models for plant disease de- tection using a comprehensive dataset of plant images. Dataset Preparation and Augmentation: Preprocess and augment the dataset by applying techniques such as rotation, zoom, and flipping to improve model performance and generalization. Performance Evaluation: Assess and compare the models based on key per- formance metrics including accuracy, loss, and confusion matrix results. Analyze Model Efficiency: Examine the accuracy trends, training efficiency, and computational requirements of CNN, AlexNet, DenseNet, and VGG16 models. Real-World Applicability: Validate the models' effectiveness in real-world scenarios by measuring their performance on test data and assessing their po- tential for early disease detection. Interpret Confusion Matrices: Analyze the confusion matrices to identify strengths and weaknesses in disease classification for each model. Discuss Model Advantages and Limitations: Highlight the benefits and limi- tations of CNN, AlexNet, DenseNet, and VGG16 models in terms of complex- ity, computational needs, and classification capabilities. IV. Methodology The methodology for this research paper involves several key steps to ensure a comprehensive evaluation of deep learning models for plant disease detection. Initially, Dataset Collection and Preparation is undertaken, where images of plant diseases are gathered and organized into categories. This is followed by Data Preprocessing, which involves techniques such as resizing, normalization, and data augmentation to enhance the quality and variability of the dataset. The next step is Model Architecture Design, where the architectures of CNN, AlexNet, DenseNet, and VGG16 are defined and implemented, specifying the layers and activation functions for each model. Model Training is then carried out, where each model is trained on the prepared dataset with carefully selected hyperparameters to optimize performance. After training, Model Evaluation is performed to assess each model's perfor- mance using metrics such as accuracy, loss, and confusion matrices. This is fol- lowed by Comparative Analysis, where the results of the different models are compared to determine their relative effectiveness in plant disease detection. Additionally, Efficiency Assessment is conducted to measure the training time and computational resources required for each model. The Practical Implemen- tation aspect evaluates how well each model can be deployed in real-world sce- narios for early disease detection. Results Interpretation and Discussion is car- ried out to explain the findings, discuss observed trends, and provide recommen- dations based on the comparative analysis of the models. V. Implementation The implementation phase begins with setting up the necessary environment, including software, libraries, and hardware needed for model development and training. Data preparation follows, where images of plant diseases are resized, normalized, and augmented to enhance the dataset's diversity and quality. Each deep learning model—CNN, AlexNet, DenseNet, and VGG16—is then built ac- cording to its specific architecture, and trained on the prepared dataset. Hyperpa- rameters are adjusted to improve performance. After training, the models are evaluated using metrics like accuracy and confu- sion matrices to determine their effectiveness. A comparative analysis identifies the best-performing model for plant disease detection. Finally, the selected model is deployed into a practical application, where it is tested in real-world conditions to ensure it works reliably for detecting plant diseases. VI. Conclusion In this research, we evaluated four deep learning models—CNN, AlexNet, DenseNet, and VGG16—for plant disease detection, all trained for 10 epochs. DenseNet emerged as the best performer with a test accuracy of 97.99%, highlighting its ability to effectively handle complex image classification tasks through its dense connections. The CNN model also performed well with 92.38% test accuracy, making it a viable option for similar tasks. VGG16 achieved moderate results with 90.37% test accuracy, likely due to overfitting. However, AlexNet significantly un- derperformed with only 18.90% accuracy, suggesting its architecture is not well-suited for this specific application. DenseNet is recommended for future work in this domain, given its superior accuracy. Declarations Author Contribution Author ContributionsA.P. and A.P. developed the study's concept, designed the methodology, and wrote the main manuscript text. P. conducted the literature review and provided expertise in deep learning model selection. M.P. and B.P. implemented the CNN, AlexNet, DenseNet, and VGG16 models, performed data preprocessing, and carried out comparative analysis. P. prepared Figures 1-12, including model performance visualizations and confusion matrices. A.P. and B.P. analyzed the results, while P. supervised the study, provided critical revisions, and finalized the manuscript. All authors reviewed and approved the final manuscript. References Diana Susan Joseph, Pranav M. Pawar, Kaustubh Chakradeo, "Real-Time Plant Disease DatasetDevelopment and Detection of Plant Disease Using Deep Learning", In IEEE Access ( Volume:12), 2024 Leninisha Shanmugam, A. L Agasta Adline, N Aishwarya, G Krithika, "Disease detection in crops using remote sensing images", In 2017 IEEE Technological Innovations in ICT for Agriculture and Rural Development (TIAR), 2017. Shalya Saxena, Sandeep Rathor, "An Ensemble-Based Model of Detecting Plant Disease using CNN and Random Forest", In 2023 6th International Conference on Information Systems and Computer Networks (ISCON), 2023. Mai Son Le, Yuei-An Liou, Minh Tuan Pham, "Crop Response to Disease and Water ScarcityQuantifiedby Normalized Difference Latent Heat Index", In IEEE Access ( Volume: 11), 2023. Achyut Morbekar, Ashi Parihar, Rashmi Jadhav, "Crop Disease Detection Using YOLO", In 2020 International Conference for Emerging Technology (INCET), 2020. S. Harika, G. Sandhyarani, D. Sagar, G.V.Subba Reddy, "Image-based Black Gram Crop Disease Detection", In 2023 International Conference on Inventive Computation Technologies (ICICT), 2023. Pallavi Pandey, Kalpesh Patyan, Manish Padekar, Rohan Mohite, Panjab Mane, Anil Avhad, "Plant Disease Detection Using Deep Learning Model -Application FarmEasy", In 2023 International Conference on Advanced Computing Technologies and Applications (ICACTA), 2023. Rahul Mishra, Dhiraj Singh, "Convolutional Neural Network Method for Effective Plant Disease Prediction", In 2023 IEEE International Conference on Integrated Circuits and Communication Systems (ICICACS), 2023. Mai Son Le, Yuei-An Liou, Minh Tuan Pham, "Crop Response to Disease and Water Scarcity Quantified by Normalized Difference Latent Heat Index", In IEEE Access ( Volume: 11), 2023. Pavan Kumar V, E Gurumohan Rao, G Anitha, G Kiran Kumar, "Plant Disease Detection using Convolutional Neural Networks", In 2021 5th International Conference on Trends in Electronics and Informatics (ICOEI), 2021. Putu Prianka Vedanty, Made Windu Antara Kesiman, Made Gede Sunarya, Gusti Ayu Agung Diatri Indradewi, "Identification of Leaf Diseases of Medicinal Plants Using K-Nearest Neighbor Based on Color, Texture, and Shape Features", In 2023 10th International Conference on Advanced Informatics: Concept, Theory and Application (ICAICTA), 2023. K. C. Deshmukh, P. A. Kulkarni, and R. S. Kumar, “Deep Learning-Based Approaches for Plant Disease Detection: A Comprehensive Review,” ScienceDirect, vol. 10, 2023. Tables Table 1: Image Distribution Across Training, Validation, and Testing Datasets Dataset Type Number of Images Training Images 7,880 Validation Images 5,625 Testing Images 5,629 Table 2: Observations and Results of CNN, AlexNet, DenseNet and VGG16 Features CNN AlexNet DenseNet VGG16 Test Accuracy 92.38% 73.46% 97.99% 90.37% Validation Accuracy 91.82% 73.30% 96.96% 90.15% Training Accuracy 90.91% 78.81% 96.13% 90.36% Table 3: Comparison of Model Parameters for CNN, AlexNet, DenseNet and VGG 16 models. As- pect CNN AlexNe t Dense- Net VGG1 6 Con- volu- tional Layers 3 Conv2D layers with 32, 64, and 128 fil- ters 5 Conv2D layers with 96, 256, 384 (twice), and 256 filters Uses dense blocks with varying fil- ter sizes within the blocks 13 Conv2D layers with vary- ing filter sizes Max- Pooling Layers 3 Max- Pool- ing2D layers 3 Max- Pooling2D layers Max- Pooling2D layers are in- cluded in the dense blocks 5 Max- Pooling2D layers Batch Normal- ization Not used Used after 1st and 2nd Conv2D layers Included in the dense blocks Not Used Drop out Lay- ers Not used 2 Drop- out layers after Dense lay- ers Not used 2 Drop- out layers Im- age Size 224x 224 224x22 4 224x224 224x22 4 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-5348075","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":376945635,"identity":"12bf74ab-e1be-4362-938d-1c4e58917849","order_by":0,"name":"Bhageerathi T","email":"data:image/png;base64,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","orcid":"","institution":"Visvesvaraya Technological University","correspondingAuthor":true,"prefix":"","firstName":"Bhageerathi","middleName":"","lastName":"T","suffix":""},{"id":376945636,"identity":"bd2e6f93-66c8-4920-b658-3063c932d853","order_by":1,"name":"Anagha M","email":"","orcid":"","institution":"Visvesvaraya Technological University","correspondingAuthor":false,"prefix":"","firstName":"Anagha","middleName":"","lastName":"M","suffix":""},{"id":376945637,"identity":"3d8db457-9aa1-4892-8f03-3189d2554d8f","order_by":2,"name":"Pushpa T S","email":"","orcid":"","institution":"Visvesvaraya Technological University","correspondingAuthor":false,"prefix":"","firstName":"Pushpa","middleName":"T","lastName":"S","suffix":""}],"badges":[],"createdAt":"2024-10-28 14:53:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5348075/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5348075/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":70579416,"identity":"6398cd86-c8e2-4210-b1ab-354bb2688749","added_by":"auto","created_at":"2024-12-04 14:54:52","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":140673,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Accuracy of CNN Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/69d38331ccfa3618d6c126c3.jpeg"},{"id":70578247,"identity":"d491fe87-9392-466b-866a-d6e9c496c2ce","added_by":"auto","created_at":"2024-12-04 14:46:52","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":129743,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Loss of CNN Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/369c6bdd772167d4e445162e.jpeg"},{"id":70578248,"identity":"e5db95d4-b363-445d-8cb4-2ff7040c4db4","added_by":"auto","created_at":"2024-12-04 14:46:52","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":116704,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Accuracy of AlexNet Model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/26d61f9322934c6dbc974aa0.jpeg"},{"id":70578254,"identity":"0ddaeadb-6a98-4cdd-92df-e0d90739021f","added_by":"auto","created_at":"2024-12-04 14:46:53","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":118422,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Loss of AlexNet Model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/50a9717d8694c481b704dbfa.jpeg"},{"id":70578250,"identity":"45c7a175-4e45-4bb5-abf4-563da373a0b8","added_by":"auto","created_at":"2024-12-04 14:46:53","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":29856,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Accuracy of DenseNet Model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/157db4e436ecf91372651d93.jpeg"},{"id":70579417,"identity":"9811e5b0-9a84-4f37-a96b-ecb906a8fc5f","added_by":"auto","created_at":"2024-12-04 14:54:53","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":108450,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Loss of DenseNet Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/afe90afe9f5eaf1f06906b18.jpeg"},{"id":70579847,"identity":"b3a1342a-1bec-45cf-bae2-e8e7431e9cac","added_by":"auto","created_at":"2024-12-04 15:02:53","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":104543,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Accuracy of VGG16 Model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/235b4550a3236e5362a1144b.jpeg"},{"id":70578251,"identity":"ff78d95b-7426-49f4-871c-59c7a05255d4","added_by":"auto","created_at":"2024-12-04 14:46:53","extension":"jpeg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":110857,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisual Representation of Training And Validation Loss of VGG16 Model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/11fd3f09ffcee05e262c6c37.jpeg"},{"id":70578255,"identity":"a6b8960a-38ea-44b2-8c0d-336fc08672f2","added_by":"auto","created_at":"2024-12-04 14:46:53","extension":"jpeg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":197834,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion Matrix – CNN Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/47d4532a80f95f75dc427a1a.jpeg"},{"id":70578258,"identity":"42e1b5d7-c8a5-4878-bcc6-255f66c98250","added_by":"auto","created_at":"2024-12-04 14:46:53","extension":"jpeg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":191844,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion Matrix – AlexNet Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage10.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/5b69957d2aaa8582e09c3b80.jpeg"},{"id":70578256,"identity":"6a73825e-4c3a-4ec3-b4e5-46849563ca5e","added_by":"auto","created_at":"2024-12-04 14:46:53","extension":"jpeg","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":214400,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion Matrix – DenseNet Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage11.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/2abe89be1f13c1477999ff52.jpeg"},{"id":70578257,"identity":"eab3b393-5777-4f6a-ada1-0cfd5f9a3a73","added_by":"auto","created_at":"2024-12-04 14:46:53","extension":"jpeg","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":198365,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion Matrix – VGG16 Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage12.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/6a166e191cf3c9f8198a0c68.jpeg"},{"id":78302612,"identity":"e19142b3-e313-4bf7-96ae-a15800a21f26","added_by":"auto","created_at":"2025-03-11 22:31:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2472214,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5348075/v1/db7072e2-dd65-4357-b969-79c3f5d182d3.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Comparative Analysis of Deep Learning Models for Plant Disease Detection","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIn the rapidly evolving field of agriculture, the health and productivity of crops are paramount to ensuring food security and economic stability. Plant diseases pose a significant threat to agricultural output, leading to substantial losses each year. Traditional methods of disease detection, which rely on visual inspection by experts, are often time-consuming, subject to human error, and not scalable across large fields. As a result, there is an increasing demand for innovative solutions that can automate and enhance the accuracy of disease detection. Deep learning has emerged as a powerful tool in the domain of computer vision, enabling the development of models that can automatically detect and classify plant diseases from leaf images. By leveraging large datasets and advanced neural network architectures, these models can identify complex patterns and subtle variations in leaf texture and color that are indicative of specific diseases.This research focuses on comparing the accuracy and performance of four widely-used deep learning models\u0026mdash;Convolutional Neural Networks (CNN), AlexNet, DenseNet, and VGG16\u0026mdash;in the context of plant disease detection. Utilizing the PlantVillage dataset, which comprises over 11,254 images of healthy and diseased leaves, the study aims to evaluate the strengths and limitations of each model. The investigation emphasizes the ability of DenseNet\u0026apos;s densely connected architecture and VGG16\u0026apos;s deep layers to manage the intricate patterns present in diseased leaves.\u003c/p\u003e"},{"header":"II.\tLiterature Review","content":"\u003cp\u003eThe paper discusses a deep learning-based model for crop disease detection us- ing the PlantVillage dataset. It compares the performance of MobileNet and In- ceptionV3 models, highlighting that InceptionV3 achieved higher accuracy in detecting crop species and diseases. The study suggests that the methodology could be extended to smartphone applications for practical use[1].\u003c/p\u003e\n\n\u003cp\u003eThe paper presents a crop disease detection system using the YOLO (You Only Look Once) algorithm, which excels in real-time image processing at 45 frames per second. By employing YOLO, which divides images into grid cells to pre- dict bounding boxes and class probabilities in one evaluation, the system offers faster and more accurate detection of plant diseases compared to traditional methods. The study highlights YOLO\u0026rsquo;s efficiency in processing leaf images to\u003c/p\u003e\n\n\n\n\n\n\u003cp\u003eidentify diseases quickly and effectively, aiming to improve crop management for farmers[2].\u003c/p\u003e\n\n\u003cp\u003eThe paper presents a comparative study of various algorithms for plant leaf dis- ease detection using deep learning techniques. It evaluates different models based on their performance in accurately classifying diseases in plant leaves. The study highlights the importance of choosing the right model for effective plant disease diagnosis, emphasizing the impact of different architectures on de- tection accuracy[3].\u003c/p\u003e\n\n\u003cp\u003eThe paper describes an automated system for detecting crop diseases using remote sensing images. The approach consists of two phases: training and mon- itoring. The training phase involves collecting images of both healthy and dis- eased crops, extracting threshold values, and using RGB layer separation. The monitoring phase uses these threshold values to compare with new images and apply techniques like Canny edge detection and histogram analysis to identify diseases. If a disease is detected, the system sends an alert to the farmer via SMS. The paper demonstrates the method\u0026apos;s effectiveness through experimental results using MATLAB[4].\u003c/p\u003e\n\n\n\u003cp\u003eThe paper introduces a lightweight CNN model for plant disease identification using Inception and Residual connections with depthwise separable convolu- tions to reduce parameters and computational complexity. The model outper- forms state-of-the-art models on three datasets, achieving high accuracy with fewer resources. This approach addresses the challenges of deploying deep learning models in resource-constrained environments[5].\u003c/p\u003e\n\n\u003cp\u003eCrop diseases and pests significantly impact agricultural productivity, mak- ing their detection crucial. This paper reviews deep learning techniques for de- tecting crop diseases and pests and proposes a Convolutional Neural Network (CNN) model for automatic diagnosis. The methodology involves image acqui- sition, preprocessing, feature extraction, and classification using CNN with transfer learning. This approach enhances detection accuracy, aiding precision agriculture[6].\u003c/p\u003e"},{"header":" III. Objectives","content":"\u003col\u003e\n \u003cli\u003eDevelop\u0026nbsp;and\u0026nbsp;Compare\u0026nbsp;Models:\u0026nbsp;Build\u0026nbsp;and\u0026nbsp;evaluate\u0026nbsp;Convolutional\u0026nbsp;Neural\u0026nbsp;Net- works (CNN), AlexNet, DenseNet, and VGG16 models for plant disease de- tection using a comprehensive dataset of plant images.\u003c/li\u003e\n \u003cli\u003eDataset Preparation and Augmentation: Preprocess and augment the dataset by\u0026nbsp;applying\u0026nbsp;techniques\u0026nbsp;such\u0026nbsp;as\u0026nbsp;rotation,\u0026nbsp;zoom,\u0026nbsp;and\u0026nbsp;flipping\u0026nbsp;to\u0026nbsp;improve\u0026nbsp;model performance and generalization.\u003c/li\u003e\n \u003cli\u003ePerformance Evaluation: Assess and compare the models based on key per- formance metrics including accuracy, loss, and confusion matrix results.\u003c/li\u003e\n \u003cli\u003eAnalyze Model Efficiency:\u0026nbsp;Examine the accuracy trends, training efficiency, and computational requirements of CNN, AlexNet, DenseNet, and VGG16 models.\u003c/li\u003e\n \u003cli\u003eReal-World Applicability: Validate the models\u0026apos; effectiveness in real-world scenarios\u0026nbsp;by\u0026nbsp;measuring\u0026nbsp;their\u0026nbsp;performance\u0026nbsp;on\u0026nbsp;test\u0026nbsp;data\u0026nbsp;and\u0026nbsp;assessing\u0026nbsp;their\u0026nbsp;po- tential for early disease detection.\u003c/li\u003e\n \u003cli\u003eInterpret Confusion Matrices: Analyze the confusion matrices to identify strengths and weaknesses in disease classification for each model.\u003c/li\u003e\n \u003cli\u003eDiscuss Model Advantages and Limitations: Highlight the benefits and limi- tations of CNN, AlexNet, DenseNet, and VGG16 models in terms of complex- ity, computational needs, and classification capabilities.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":" IV. Methodology","content":"\u003cp\u003eThe methodology for this research paper involves several key steps to ensure a comprehensive evaluation of deep learning models for plant disease detection. Initially, Dataset Collection and Preparation is undertaken, where images of plant diseases are gathered and organized into categories. This is followed by Data Preprocessing, which involves techniques such as resizing, normalization, and data augmentation to enhance the quality and variability of the dataset.\u003c/p\u003e\n\u003cp\u003eThe next step is Model Architecture Design, where the architectures of CNN, AlexNet, DenseNet, and VGG16 are defined and implemented, specifying the layers and activation functions for each model. Model Training is then carried out,\u0026nbsp;where\u0026nbsp;each\u0026nbsp;model\u0026nbsp;is\u0026nbsp;trained\u0026nbsp;on\u0026nbsp;the\u0026nbsp;prepared\u0026nbsp;dataset\u0026nbsp;with\u0026nbsp;carefully\u0026nbsp;selected hyperparameters to optimize performance.\u003c/p\u003e\n\u003cp\u003eAfter training, Model Evaluation is performed to assess each model\u0026apos;s perfor- mance using metrics such as accuracy, loss, and confusion matrices. This is fol- lowed by Comparative Analysis, where the results of the different models are compared to determine their relative effectiveness in plant disease detection. Additionally, Efficiency Assessment is conducted to measure the training time and computational resources required for each model. The Practical Implemen- tation aspect evaluates how well each model can be deployed in real-world sce- narios for early disease detection. Results Interpretation and Discussion is car- ried out to explain the findings, discuss observed trends, and provide recommen- dations based on the comparative analysis of the models.\u003c/p\u003e"},{"header":"V.\tImplementation","content":"\u003cp\u003eThe implementation phase begins with setting up the necessary environment, including software, libraries, and hardware needed for model development and training. Data preparation follows, where images of plant diseases are resized, normalized, and augmented to enhance the dataset\u0026apos;s diversity and quality. Each deep learning model\u0026mdash;CNN, AlexNet, DenseNet, and VGG16\u0026mdash;is then built ac- cording to its specific architecture, and trained on the prepared dataset. Hyperpa- rameters are adjusted to improve performance.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAfter training, the models are evaluated using metrics like accuracy and confu- sion matrices to determine their effectiveness. A comparative analysis identifies the best-performing model for plant disease detection. Finally, the selected model is deployed into a practical application, where it is tested in real-world conditions to ensure it works reliably for detecting plant diseases.\u003c/p\u003e"},{"header":"VI. Conclusion","content":"\u003cp\u003eIn this research, we evaluated four deep learning models\u0026mdash;CNN, AlexNet, DenseNet, and VGG16\u0026mdash;for plant disease detection, all trained for 10 epochs. DenseNet emerged as the best performer with a test accuracy of 97.99%, highlighting its ability to effectively handle complex image classification tasks through its dense connections. The CNN model also performed well with 92.38% test accuracy, making it a viable option for similar tasks. VGG16 achieved moderate results with 90.37% test accuracy, likely due to overfitting. However, AlexNet significantly un- derperformed with only 18.90% accuracy, suggesting its architecture is not well-suited for this specific application. DenseNet is recommended for future work in this domain, given its superior accuracy.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAuthor ContributionsA.P. and A.P. developed the study's concept, designed the methodology, and wrote the main manuscript text. P. conducted the literature review and provided expertise in deep learning model selection. M.P. and B.P. implemented the CNN, AlexNet, DenseNet, and VGG16 models, performed data preprocessing, and carried out comparative analysis. P. prepared Figures 1-12, including model performance visualizations and confusion matrices. A.P. and B.P. analyzed the results, while P. supervised the study, provided critical revisions, and finalized the manuscript. All authors reviewed and approved the final manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDiana Susan Joseph, Pranav M. Pawar, Kaustubh Chakradeo, \u0026quot;Real-Time Plant Disease DatasetDevelopment and Detection of Plant Disease Using Deep Learning\u0026quot;, In IEEE Access ( Volume:12), 2024\u003c/li\u003e\n\u003cli\u003eLeninisha Shanmugam, A. L Agasta Adline, N Aishwarya, G Krithika, \u0026quot;Disease detection in crops using remote sensing images\u0026quot;, In 2017 IEEE Technological Innovations in ICT for Agriculture and Rural Development (TIAR), 2017.\u003c/li\u003e\n\u003cli\u003eShalya Saxena, Sandeep Rathor, \u0026quot;An Ensemble-Based Model of Detecting Plant Disease using CNN and Random Forest\u0026quot;, In 2023 6th International Conference on Information Systems and Computer Networks (ISCON), 2023.\u003c/li\u003e\n\u003cli\u003eMai Son Le, Yuei-An Liou, Minh Tuan Pham, \u0026quot;Crop Response to Disease and Water ScarcityQuantifiedby Normalized Difference Latent Heat Index\u0026quot;, In IEEE Access ( Volume: 11), 2023.\u003c/li\u003e\n\u003cli\u003eAchyut Morbekar, Ashi Parihar, Rashmi Jadhav, \u0026quot;Crop Disease Detection Using YOLO\u0026quot;, In 2020 International Conference for Emerging Technology (INCET), 2020.\u003c/li\u003e\n\u003cli\u003eS. Harika, G. Sandhyarani, D. Sagar, G.V.Subba Reddy, \u0026quot;Image-based Black Gram Crop Disease Detection\u0026quot;, In 2023 International Conference on Inventive Computation Technologies (ICICT), 2023.\u003c/li\u003e\n\u003cli\u003ePallavi Pandey, Kalpesh Patyan, Manish Padekar, Rohan Mohite, Panjab Mane, Anil Avhad, \u0026quot;Plant Disease Detection Using Deep Learning Model -Application FarmEasy\u0026quot;, In 2023 International Conference on Advanced Computing Technologies and Applications (ICACTA), 2023.\u003c/li\u003e\n\u003cli\u003eRahul Mishra, Dhiraj Singh, \u0026quot;Convolutional Neural Network Method for Effective Plant Disease Prediction\u0026quot;, In 2023 IEEE International Conference on Integrated Circuits and Communication Systems (ICICACS), 2023.\u003c/li\u003e\n\u003cli\u003eMai Son Le, Yuei-An Liou, Minh Tuan Pham, \u0026quot;Crop Response to Disease and Water Scarcity Quantified by Normalized Difference Latent Heat Index\u0026quot;, In IEEE Access ( Volume: 11), 2023.\u003c/li\u003e\n\u003cli\u003ePavan Kumar V, E Gurumohan Rao, G Anitha, G Kiran Kumar, \u0026quot;Plant Disease Detection using Convolutional Neural Networks\u0026quot;, In 2021 5th International Conference on Trends in Electronics and Informatics (ICOEI), 2021.\u003c/li\u003e\n\u003cli\u003ePutu Prianka Vedanty, Made Windu Antara Kesiman, Made Gede Sunarya, Gusti Ayu Agung Diatri Indradewi, \u0026quot;Identification of Leaf Diseases of Medicinal Plants Using K-Nearest Neighbor Based on Color, Texture, and Shape Features\u0026quot;, In 2023 10th International Conference on Advanced Informatics: Concept, Theory and Application (ICAICTA), 2023.\u003c/li\u003e\n\u003cli\u003eK. C. Deshmukh, P. A. Kulkarni, and R. S. Kumar, \u0026ldquo;Deep Learning-Based Approaches for Plant Disease Detection: A Comprehensive Review,\u0026rdquo; ScienceDirect, vol. 10, 2023.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u0026nbsp;\u003cstrong\u003eTable 1: Image Distribution Across Training, Validation, and Testing Datasets\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 51.5707%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDataset\u0026nbsp;Type\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 48.4293%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNumber\u0026nbsp;of\u0026nbsp;Images\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 51.5707%;\"\u003e\n \u003cp\u003eTraining\u0026nbsp;Images\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 48.4293%;\"\u003e\n \u003cp\u003e7,880\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 51.5707%;\"\u003e\n \u003cp\u003eValidation\u0026nbsp;Images\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 48.4293%;\"\u003e\n \u003cp\u003e5,625\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 51.5707%;\"\u003e\n \u003cp\u003eTesting\u0026nbsp;Images\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 48.4293%;\"\u003e\n \u003cp\u003e5,629\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\u003cp\u003e\u003cstrong\u003eTable 2: Observations and Results of CNN, AlexNet, DenseNet and VGG16\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30.9859%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFeatures\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.493%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCNN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAlexNet\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDenseNet\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVGG16\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30.9859%;\"\u003e\n \u003cp\u003eTest\u0026nbsp;Accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.493%;\"\u003e\n \u003cp\u003e92.38%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e73.46%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e97.99%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e90.37%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30.9859%;\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.493%;\"\u003e\n \u003cp\u003e91.82%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e73.30%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e96.96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e90.15%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 30.9859%;\"\u003e\n \u003cp\u003eTraining\u0026nbsp;Accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.493%;\"\u003e\n \u003cp\u003e90.91%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e78.81%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e96.13%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.8404%;\"\u003e\n \u003cp\u003e90.36%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3: Comparison of Model Parameters for CNN, AlexNet, DenseNet and VGG 16 models.\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAs- pect\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCNN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.7317%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAlexNe\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003et\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23.4756%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDense-\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eNet\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.4268%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVGG1\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e6\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCon-\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003evolu- tional Layers\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e3\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eConv2D layers\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003ewith 32,\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e64, and\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e128\u0026nbsp;fil- ters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.7317%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e5\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eConv2D layers\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003ewith 96,\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e256, 384\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(twice),\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eand\u0026nbsp;256 filters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23.4756%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUses\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003edense\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eblocks\u0026nbsp;with varying\u0026nbsp;fil- ter sizes within the blocks\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.4268%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e13\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eConv2D layers\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003ewith\u0026nbsp;vary- ing filter sizes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003eMax-\u0026nbsp;Pooling Layers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003cp\u003eMax-\u0026nbsp;Pool- ing2D layers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.7317%;\"\u003e\n \u003cp\u003e3\u0026nbsp;Max-\u003c/p\u003e\n \u003cp\u003ePooling2D layers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23.4756%;\"\u003e\n \u003cp\u003eMax-\u0026nbsp;Pooling2D\u0026nbsp;layers\u0026nbsp;are\u0026nbsp;in- cluded\u0026nbsp;in\u0026nbsp;the dense\u0026nbsp;blocks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.4268%;\"\u003e\n \u003cp\u003e5\u0026nbsp;Max-\u003c/p\u003e\n \u003cp\u003ePooling2D layers\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003eBatch Normal- ization\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003eNot used\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.7317%;\"\u003e\n \u003cp\u003eUsed\u0026nbsp;after\u0026nbsp;1st and\u0026nbsp;2nd Conv2D\u003c/p\u003e\n \u003cp\u003elayers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23.4756%;\"\u003e\n \u003cp\u003eIncluded\u0026nbsp;in\u0026nbsp;the\u0026nbsp;dense\u003c/p\u003e\n \u003cp\u003eblocks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.4268%;\"\u003e\n \u003cp\u003eNot Used\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003eDrop\u0026nbsp;out Lay-\u003c/p\u003e\n \u003cp\u003eers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003eNot used\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.7317%;\"\u003e\n \u003cp\u003e2\u0026nbsp;Drop- out layers\u003c/p\u003e\n \u003cp\u003eafter\u0026nbsp;Dense\u0026nbsp;lay- ers\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23.4756%;\"\u003e\n \u003cp\u003eNot used\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.4268%;\"\u003e\n \u003cp\u003e2\u0026nbsp;Drop- out layers\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003eIm-\u0026nbsp;age Size\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6829%;\"\u003e\n \u003cp\u003e224x\u003c/p\u003e\n \u003cp\u003e224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.7317%;\"\u003e\n \u003cp\u003e224x22\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23.4756%;\"\u003e\n \u003cp\u003e224x224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.4268%;\"\u003e\n \u003cp\u003e224x22\u003c/p\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Plant Disease Detection, Deep Learning, Convolutional Neural Networks (CNN), AlexNet, DenseNet, VGG16, Image Classification, PlantVillage Dataset, Precision Agriculture, Automated Diagnosis, Model Comparison, Accuracy","lastPublishedDoi":"10.21203/rs.3.rs-5348075/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5348075/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe detection of plant diseases through deep learning models repre- sents a significant advancement in agricultural management. This study pro- vides a comprehensive accuracy comparison of four prominent deep learning models\u0026mdash;Convolutional Neural Networks (CNN), AlexNet, DenseNet, and VGG16\u0026mdash;for identifying plant diseases from leaf images. Leveraging the PlantVillage dataset, which includes over 11,254 images of healthy and dis- eased leaves, the research investigates the strengths and limitations of each model in terms of accuracy, feature extraction, and classification performance. DenseNet's densely connected architecture and VGG16's deep layers are high- lighted for their superior ability to handle complex patterns in diseased leaves. The study demonstrates that DenseNet achieves the highest accuracy, making it a viable solution for real-time disease detection in precision agriculture. By comparing these models, the research aims to guide the selection of the most effective deep learning approach for improving plant health monitoring.\u003c/p\u003e","manuscriptTitle":"Comparative Analysis of Deep Learning Models for Plant Disease Detection","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-04 14:46:48","doi":"10.21203/rs.3.rs-5348075/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"89e7ba7c-090f-472a-86f2-80cc7dfc51f9","owner":[],"postedDate":"December 4th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-03-11T22:23:15+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-04 14:46:48","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5348075","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5348075","identity":"rs-5348075","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}