AI-Powered System for Detecting and Classifying Plant Diseases using Image Processing Techniques

AI-Powered System for Detecting and Classifying Plant Diseases using Image Processing Techniques | 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 AI-Powered System for Detecting and Classifying Plant Diseases using Image Processing Techniques Dr Ramu Vankudoth This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7218586/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 This paper presents a novel approach to automated plant disease detection and classification using advanced image processing and deep learning techniques. Early detection of plant diseases is crucial for sustainable agricultural practices and food security. Our proposed system leverages convolutional neural networks (CNNs) to analyze leaf images and accurately identify various plant diseases across multiple crop species. The methodology includes image preprocessing, segmentation, feature extraction, and classification using a custom CNN architecture. The system was trained and validated on a diverse dataset containing 38,000 images spanning 14 crop species and 26 diseases. Experimental results demonstrate 97.89% classification accuracy, outperforming existing methods. The system is implemented as a lightweight mobile application allowing farmers to diagnose plant diseases in real-time using only a smartphone camera, potentially reducing crop losses and pesticide usage through early intervention. This research contributes to precision agriculture by providing an accessible, cost-effective tool for disease management in both developed and developing agricultural contexts. Computer Architecture and Engineering Plant disease detection Convolutional neural networks Image processing Deep learning Precision agriculture Mobile applications Figures Figure 1 Figure 2 1. Introduction Plant diseases pose a significant threat to global food security and agricultural sustainability, with annual crop losses estimated between 20-40% due to pathogen infections (Strange and Scott, 2005). Early and accurate detection of plant diseases remains challenging, particularly in developing regions where agricultural expertise is limited. Traditional disease diagnosis relies heavily on human experts, whose availability is often constrained by geographical and economic factors. Recent advances in artificial intelligence, particularly in computer vision and deep learning, offer promising solutions for automated plant disease detection. These technologies can potentially transform disease management practices by providing timely, accurate, and accessible diagnostic tools to farmers worldwide (Mohanty et al., 2016). This paper presents a comprehensive AI-powered system for detecting and classifying plant diseases using image processing techniques. The proposed approach combines advanced image preprocessing methods with state-of-the-art convolutional neural networks to analyze leaf images and identify diseases with high accuracy. The system is designed to be computationally efficient for deployment on mobile devices, making it accessible to farmers in diverse agricultural settings. The primary contributions of this research include: A novel image preprocessing pipeline optimized for plant leaf analysis in varied lighting and background conditions A custom CNN architecture designed specifically for plant disease classification with improved feature extraction capabilities An extensive evaluation on a diverse, real-world dataset encompassing multiple crop species and disease categories Implementation of a lightweight mobile application that provides real-time disease diagnosis and treatment recommendations Comparative analysis against existing approaches demonstrating superior performance in accuracy and computational efficiency 2. RELATED WORK Research in automated plant disease detection has evolved significantly over the past decade. Early approaches relied on traditional image processing techniques and conventional machine learning methods. Barbedo (2013) surveyed various methods for plant disease identification based on digital image processing, highlighting challenges related to image acquisition and feature extraction. With the advent of deep learning, several researchers have explored CNN-based approaches for plant disease classification. Sladojevic et al. (2016) implemented a CNN model to recognize 13 different plant diseases with an accuracy of 96.3%. Mohanty et al. (2016) used AlexNet and GoogLeNet architectures on the PlantVillage dataset, achieving accuracies up to 99.35% under controlled conditions, though performance decreased significantly when tested on images collected in real-field environments. More recent studies have addressed the challenge of real-world conditions. Ferentinos (2018) developed deep learning models trained on an open database of 87,848 images, achieving a 99.53% success rate for plant disease detection. Fuentes et al. (2017) proposed a deep-learning-based detector for real-time tomato disease and pest recognition, using the Faster Region-based CNN (Faster R-CNN) framework. Mobile applications for plant disease diagnosis have also emerged. Ramcharan et al. (2017) developed a cassava disease detection system using the transfer learning approach with the TensorFlow platform. Similarly, Johannes et al. (2017) created a mobile app for automatic plant disease diagnosis. Despite these advances, challenges remain in developing systems that maintain high accuracy across diverse environmental conditions while being computationally efficient enough for mobile deployment. Our work addresses these gaps by introducing novel preprocessing techniques and a custom CNN architecture optimized for resource-constrained environments. 3. PORPOSED MODELLING 3.1 Dataset Acquisition and Preparation Our dataset consists of 38,000 images representing 14 crop species and 26 different diseases, as well as healthy specimens. The images were collected from: Public repositories: PlantVillage dataset (Hughes and Salathé, 2015) Field surveys: Images captured using digital cameras and smartphones in various agricultural settings across different geographical regions Agricultural research institutions: Curated and labeled images provided by partnering universities and research centers The dataset encompasses major crop species including rice, wheat, maize, potato, tomato, apple, grape, and citrus. For each species, multiple disease categories were included, representing fungal, bacterial, and viral infections. Healthy plant samples were also incorporated to enable the system to distinguish between diseased and non-diseased states. To ensure diversity, images were collected under varying conditions: Different lighting conditions (natural sunlight, shade, indoor lighting) Various backgrounds (soil, grass, indoor surfaces) Different perspectives and distances Various stages of disease progression Different leaf positions and orientations The collected images were manually verified and labeled by agricultural experts to ensure accuracy. The dataset was randomly split into training (70%), validation (15%), and testing (15%) sets, maintaining class distribution across all sets. 3.2 Image Preprocessing A robust preprocessing pipeline was developed to normalize the images and enhance disease-specific features: Resizing and standardization: All images were resized to 256×256 pixels and standardized to RGB color format. Background removal: A combination of GrabCut algorithm (Rother et al., 2004) and color-based segmentation was used to isolate leaf regions from the background. Illumination normalization: A contrast limited adaptive histogram equalization (CLAHE) technique was applied to compensate for variable lighting conditions. Data augmentation: To improve model robustness, the training dataset was augmented using: Random rotations (±30°) Horizontal and vertical flips Random brightness and contrast adjustments (±10%) Slight zoom variations (0.9-1.1x) Random cropping (maintaining at least 80% of the original content) Color space transformation: Images were analyzed in multiple color spaces (RGB, HSV, and Lab*) to extract complementary features, particularly for diseases that manifest as color abnormalities. 3.3 Feature Extraction and Selection Both traditional image processing features and CNN-learned features were utilized: Traditional features: Color features: Color histograms, color moments, and color coherence vectors Texture features: Gray Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LBP) Shape features: Hu moments, Fourier descriptors CNN-based feature extraction: A custom CNN architecture was designed to automatically learn hierarchical features from the preprocessed images. Feature selection was performed using a combination of Principal Component Analysis (PCA) for dimensional reduction of traditional features and attention mechanisms within the CNN to focus on disease-relevant image regions. 3.4 CNN Architecture and Model Development We developed a custom CNN architecture optimized for plant disease classification: Additionally, we incorporated: Attention mechanism: A spatial attention module was introduced after the third convolutional block to help the network focus on disease-specific regions. Residual connections: Skip connections were added to facilitate gradient flow during training and enable deeper network training. Transfer learning: For comparison, we fine-tuned pre-trained models (ResNet-50, MobileNetV2, and EfficientNet-B0) on our dataset. The model was trained using: Loss function: Categorical cross-entropy Optimizer: Adam with learning rate of 0.0001 Batch size: 32 Early stopping: Patience of 15 epochs monitoring validation loss Learning rate reduction: By factor of 0.2 when validation loss plateaued 3.5 Model Deployment The trained model was optimized for mobile deployment through: Model quantization: Weights were quantized to 8-bit integers, reducing model size by approximately 75%. Pruning: Non-essential connections were pruned, further reducing computational requirements. TensorFlow Lite conversion: The optimized model was converted to TensorFlow Lite format for efficient mobile execution. A cross-platform mobile application was developed with the following features: Real-time image capture and analysis Offline operation capability Disease information and treatment recommendations Historical tracking of detections Integration with agricultural extension services 4. RESULTS AND DISCUSSIONS In this section all the results and the discussions should be made. 4.1 Performance Evaluation The performance of our system was evaluated using standard metrics: Accuracy Proportion of correctly classified instances Precision True positives divided by predicted positives Recall True positives divided by actual positives F1-score Harmonic mean of precision and recall Table 1 presents the overall performance of different models tested: Table 1 Performance comparison of different models Model Accuracy (%) Precision (%) Recall (%) F1-score (%) CNN Proposed 97.89 97.65 97.72 97.68 ResNet-50 96.54 96.38 96.41 96.39 MobileNetV2 95.82 95.66 95.71 95.68 EfficientNet-B0 96.93 96.81 96.75 96.78 Traditional ML (SVM with HOG features) 84.21 83.95 84.17 84.06 Our custom CNN architecture achieved the highest performance across all metrics, with an overall accuracy of 97.89%. The confusion matrix analysis revealed that most misclassifications occurred between visually similar diseases affecting the same plant species. 4.2 Impact of Preprocessing Steps To evaluate the contribution of different preprocessing steps, ablation studies were conducted. Table 2 shows the impact of each preprocessing component: Table 2 Impact of preprocessing steps on model accuracy Preprocessing Configuration Accuracy (%) Complete pipeline 97.89 Without background removal 95.37 Without illumination normalization 96.12 Without data augmentation 94.86 Without color space transformation 97.04 Basic preprocessing only (resize + normalize) 92.55 The results demonstrate that each preprocessing component contributes to the overall performance, with data augmentation and background removal showing the most significant impact. 4.3 Performance across Different Environmental Conditions The system's robustness was tested across various environmental conditions. Table 3 summarizes the accuracy across different imaging scenarios: Table 3 Accuracy under different environmental conditions Condition Accuracy (%) Controlled environment (lab setting) 99.23 Natural outdoor lighting 96.82 Low light conditions 95.41 Varying backgrounds 96.25 Different distances 97.14 Early-stage disease symptoms 93.87 While performance remained high across most conditions, early-stage disease detection presented the greatest challenge, as expected. This highlights an area for future improvement. 4.4 Computational Efficiency and Mobile Performance The optimized model demonstrated excellent performance on mobile devices: Table 4 Mobile performance metrics Metric Value Model size 8.7 MB Average inference time (mid-range smartphone) 312 ms Memory usage 145 MB Battery consumption (per 100 inferences) ~ 1% These results confirm the system's suitability for deployment on resource-constrained devices, making it accessible to farmers with basic smartphone hardware. 4.5 Comparison with Existing Systems Our system was compared with other published plant disease detection approaches: Table 5 Comparison with existing approaches Approach Dataset Size Number of Classes Accuracy (%) Mobile Compatible Proposed 38,000 40 97.89 Yes Mohanty et al. ( 2016 ) 54,306 38 99.35* No Ferentinos ( 2018 ) 87,848 58 99.53* No Ramcharan et al. ( 2017 ) 2,756 5 93.00 Yes Too et al. ( 2019 ) 54,306 38 98.00 No *Accuracy on controlled environment images only. Performance drops significantly (15–30%) on real-field images. While some approaches report marginally higher accuracy on controlled datasets, our system maintains high performance across real-world conditions while being optimized for mobile deployment. 5. Field Application and User Study A pilot study was conducted with 50 farmers across different agricultural regions to evaluate the practical utility of the mobile application. Participants used the application for three months during a growing season. Key findings include: 94% of participants reported the application was "easy" or "very easy" to use Disease identification by the app matched expert diagnosis in 92% of cases 87% of farmers reported earlier disease detection compared to their usual practices 76% reported reduced pesticide usage due to more targeted and timely interventions Average estimated crop loss reduction was 23% compared to previous seasons Qualitative feedback highlighted the value of offline functionality and the integrated treatment recommendations. 5. CONCLUSION This paper presented a comprehensive AI-powered system for detecting and classifying plant diseases using image processing and deep learning techniques. Our approach combines robust preprocessing with a custom CNN architecture optimized for mobile deployment. The system achieves 97.89% classification accuracy across a diverse dataset of 38,000 images spanning 14 crop species and 26 diseases. The field application study demonstrates the practical utility of the system, with farmers reporting earlier disease detection, reduced pesticide usage, and decreased crop losses. The mobile implementation makes advanced disease diagnostic technology accessible to farmers in diverse agricultural settings, including resource-limited regions. Future work will focus on: Expanding the disease database to include more crop species and disease categories Improving early-stage disease detection through temporal analysis of plant development Incorporating environmental data (temperature, humidity, soil conditions) to enhance diagnostic accuracy Developing region-specific models that account for local disease prevalence and manifestation Implementing cloud synchronization for continuous model improvement through federated learning The system presented in this paper contributes to sustainable agricultural practices by providing a cost-effective tool for early plant disease detection, potentially reducing crop losses and environmental impact of excessive pesticide use. References Barbedo, J.G.A. (2013). Digital image processing techniques for detecting, quantifying and classifying plant diseases. SpringerPlus, 2(1), 660. Ferentinos, K.P. (2018). Deep learning models for plant disease detection and diagnosis. Computers and Electronics in Agriculture, 145, 311-318. Fuentes, A., Yoon, S., Kim, S.C., & Park, D.S. (2017). A robust deep-learning-based detector for real-time tomato plant diseases and pests recognition. Sensors, 17(9), 2022. Hughes, D.P., & Salathé, M. (2015). An open access repository of images on plant health to enable the development of mobile disease diagnostics. arXiv preprint arXiv:1511.08060. Johannes, A., Picon, A., Alvarez-Gila, A., Echazarra, J., Rodriguez-Vaamonde, S., Navajas, A.D., & Ortiz-Barredo, A. (2017). Automatic plant disease diagnosis using mobile capture devices, applied on a wheat use case. Computers and Electronics in Agriculture, 138, 200-209. Mohanty, S.P., Hughes, D.P., & Salathé, M. (2016). Using deep learning for image-based plant disease detection. Frontiers in Plant Science, 7, 1419. Ramcharan, A., Baranowski, K., McCloskey, P., Ahmed, B., Legg, J., & Hughes, D.P. (2017). Deep learning for image-based cassava disease detection. Frontiers in Plant Science, 8, 1852. Rother, C., Kolmogorov, V., & Blake, A. (2004). GrabCut: Interactive foreground extraction using iterated graph cuts. ACM Transactions on Graphics, 23(3), 309-314. Sladojevic, S., Arsenovic, M., Anderla, A., Culibrk, D., & Stefanovic, D. (2016). Deep neural networks based recognition of plant diseases by leaf image classification. Computational Intelligence and Neuroscience, 2016, 3289801. Strange, R.N., & Scott, P.R. (2005). Plant disease: A threat to global food security. Annual Review of Phytopathology, 43, 83-116. Too, E.C., Yujian, L., Njuki, S., & Yingchun, L. (2019). A comparative study of fine-tuning deep learning models for plant disease identification. Computers and Electronics in Agriculture, 161, 272-279. Additional Declarations The authors declare potential competing interests as follows: 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-7218586","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":491131193,"identity":"09dae61e-bb64-4dfb-9ac8-c785ec8ca39b","order_by":0,"name":"Dr Ramu Vankudoth","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYLACHgYbJDaRWtJI13KYBDeZt59Ok3jz53y0eXvzA4YfNQwy5oS0yJzJ3SY5t+127pwzxwwYe44x8Fg2ENAiwZC7TZq34XbuDIkcBgbeBgYegwOEtPC/3SbN8+dc7gz5NwyMf4nSIgG0hYftANAWHgZm4myReLvZcm5bcu4MnjSDwzLHJIhxWO7GG2/+2OXOYD/88OGbGht7glpQwAFQcIyCUTAKRsEooAIAAAdPOiDDSaO9AAAAAElFTkSuQmCC","orcid":"","institution":"Malla Reddy Deemed to be University","correspondingAuthor":true,"prefix":"Dr","firstName":"Ramu","middleName":"","lastName":"Vankudoth","suffix":""}],"badges":[],"createdAt":"2025-07-26 05:23:14","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":true,"conflictsOfInterestStatement":true,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":true,"humanSubjectCaseReport":true,"vertebrateSubjectEthicalGuidelines":true},"doi":"10.21203/rs.3.rs-7218586/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7218586/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87898631,"identity":"16f83a2a-ef0a-4a0f-97b8-d2334e726f32","added_by":"auto","created_at":"2025-07-30 07:54:28","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1380343,"visible":true,"origin":"","legend":"\u003cp\u003eUnnumbered image in the Porposed Modelling section.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7218586/v1/e7d5292b7db71bf561d456c1.png"},{"id":87898634,"identity":"2c4c0fcd-37b2-4ccf-9bc0-6dff846f89e5","added_by":"auto","created_at":"2025-07-30 07:54:29","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1130705,"visible":true,"origin":"","legend":"\u003cp\u003eUnnumbered image in the Porposed Modelling section.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7218586/v1/d735ce5012c64eb9fe489829.png"},{"id":87899348,"identity":"775ac251-a39f-4378-9938-34cd5a5b88da","added_by":"auto","created_at":"2025-07-30 08:02:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3769835,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7218586/v1/651b15f1-21f9-4e46-bd10-cd7d838a9507.pdf"}],"financialInterests":"The authors declare potential competing interests as follows: ","formattedTitle":"\u003cp\u003eAI-Powered System for Detecting and Classifying Plant Diseases using Image Processing Techniques\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003ePlant diseases pose a significant threat to global food security and agricultural sustainability, with annual crop losses estimated between 20-40% due to pathogen infections (Strange and Scott, 2005). Early and accurate detection of plant diseases remains challenging, particularly in developing regions where agricultural expertise is limited. Traditional disease diagnosis relies heavily on human experts, whose availability is often constrained by geographical and economic factors.\u003c/p\u003e\n\u003cp\u003eRecent advances in artificial intelligence, particularly in computer vision and deep learning, offer promising solutions for automated plant disease detection. These technologies can potentially transform disease management practices by providing timely, accurate, and accessible diagnostic tools to farmers worldwide (Mohanty et al., 2016).\u003c/p\u003e\n\u003cp\u003eThis paper presents a comprehensive AI-powered system for detecting and classifying plant diseases using image processing techniques. The proposed approach combines advanced image preprocessing methods with state-of-the-art convolutional neural networks to analyze leaf images and identify diseases with high accuracy. The system is designed to be computationally efficient for deployment on mobile devices, making it accessible to farmers in diverse agricultural settings.\u003c/p\u003e\n\u003cp\u003eThe primary contributions of this research include:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eA novel image preprocessing pipeline optimized for plant leaf analysis in varied lighting and background conditions\u003c/li\u003e\n \u003cli\u003eA custom CNN architecture designed specifically for plant disease classification with improved feature extraction capabilities\u003c/li\u003e\n \u003cli\u003eAn extensive evaluation on a diverse, real-world dataset encompassing multiple crop species and disease categories\u003c/li\u003e\n \u003cli\u003eImplementation of a lightweight mobile application that provides real-time disease diagnosis and treatment recommendations\u003c/li\u003e\n \u003cli\u003eComparative analysis against existing approaches demonstrating superior performance in accuracy and computational efficiency\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"2.\tRELATED WORK","content":"\u003cp\u003eResearch in automated plant disease detection has evolved significantly over the past decade. Early approaches relied on traditional image processing techniques and conventional machine learning methods. Barbedo (2013) surveyed various methods for plant disease identification based on digital image processing, highlighting challenges related to image acquisition and feature extraction.\u003c/p\u003e\n\u003cp\u003eWith the advent of deep learning, several researchers have explored CNN-based approaches for plant disease classification. Sladojevic et al. (2016) implemented a CNN model to recognize 13 different plant diseases with an accuracy of 96.3%. Mohanty et al. (2016) used AlexNet and GoogLeNet architectures on the PlantVillage dataset, achieving accuracies up to 99.35% under controlled conditions, though performance decreased significantly when tested on images collected in real-field environments.\u003c/p\u003e\n\u003cp\u003eMore recent studies have addressed the challenge of real-world conditions. Ferentinos (2018) developed deep learning models trained on an open database of 87,848 images, achieving a 99.53% success rate for plant disease detection. Fuentes et al. (2017) proposed a deep-learning-based detector for real-time tomato disease and pest recognition, using the Faster Region-based CNN (Faster R-CNN) framework.\u003c/p\u003e\n\u003cp\u003eMobile applications for plant disease diagnosis have also emerged. Ramcharan et al. (2017) developed a cassava disease detection system using the transfer learning approach with the TensorFlow platform. Similarly, Johannes et al. (2017) created a mobile app for automatic plant disease diagnosis.\u003c/p\u003e\n\u003cp\u003eDespite these advances, challenges remain in developing systems that maintain high accuracy across diverse environmental conditions while being computationally efficient enough for mobile deployment. Our work addresses these gaps by introducing novel preprocessing techniques and a custom CNN architecture optimized for resource-constrained environments.\u003c/p\u003e"},{"header":"3.\tPORPOSED MODELLING ","content":"\u003cp\u003e3.1 Dataset Acquisition and Preparation\u003c/p\u003e\n\u003cp\u003eOur dataset consists of 38,000 images representing 14 crop species and 26 different diseases, as well as healthy specimens. The images were collected from:\u003c/p\u003e\n\u003cp\u003ePublic repositories: PlantVillage dataset (Hughes and Salath\u0026eacute;, 2015)\u003c/p\u003e\n\u003cp\u003eField surveys: Images captured using digital cameras and smartphones in various agricultural settings across different geographical regions\u003c/p\u003e\n\u003cp\u003eAgricultural research institutions: Curated and labeled images provided by partnering universities and research centers\u003c/p\u003e\n\u003cp\u003eThe dataset encompasses major crop species including rice, wheat, maize, potato, tomato, apple, grape, and citrus. For each species, multiple disease categories were included, representing fungal, bacterial, and viral infections. Healthy plant samples were also incorporated to enable the system to distinguish between diseased and non-diseased states.\u003c/p\u003e\n\u003cp\u003eTo ensure diversity, images were collected under varying conditions:\u003c/p\u003e\n\u003cp\u003eDifferent lighting conditions (natural sunlight, shade, indoor lighting)\u003c/p\u003e\n\u003cp\u003eVarious backgrounds (soil, grass, indoor surfaces)\u003c/p\u003e\n\u003cp\u003eDifferent perspectives and distances\u003c/p\u003e\n\u003cp\u003eVarious stages of disease progression\u003c/p\u003e\n\u003cp\u003eDifferent leaf positions and orientations\u003c/p\u003e\n\u003cp\u003eThe collected images were manually verified and labeled by agricultural experts to ensure accuracy. The dataset was randomly split into training (70%), validation (15%), and testing (15%) sets, maintaining class distribution across all sets.\u003c/p\u003e\n\u003cp\u003e3.2 Image Preprocessing\u003c/p\u003e\n\u003cp\u003eA robust preprocessing pipeline was developed to normalize the images and enhance disease-specific features:\u003c/p\u003e\n\u003cp\u003eResizing and standardization: All images were resized to 256\u0026times;256 pixels and standardized to RGB color format.\u003c/p\u003e\n\u003cp\u003eBackground removal: A combination of GrabCut algorithm (Rother et al., 2004) and color-based segmentation was used to isolate leaf regions from the background.\u003c/p\u003e\n\u003cp\u003eIllumination normalization: A contrast limited adaptive histogram equalization (CLAHE) technique was applied to compensate for variable lighting conditions.\u003c/p\u003e\n\u003cp\u003eData augmentation: To improve model robustness, the training dataset was augmented using:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRandom rotations (\u0026plusmn;30\u0026deg;)\u003c/p\u003e\n\u003cp\u003eHorizontal and vertical flips\u003c/p\u003e\n\u003cp\u003eRandom brightness and contrast adjustments (\u0026plusmn;10%)\u003c/p\u003e\n\u003cp\u003eSlight zoom variations (0.9-1.1x)\u003c/p\u003e\n\u003cp\u003eRandom cropping (maintaining at least 80% of the original content)\u003c/p\u003e\n\u003cp\u003eColor space transformation: Images were analyzed in multiple color spaces (RGB, HSV, and Lab*) to extract complementary features, particularly for diseases that manifest as color abnormalities.\u003c/p\u003e\n\u003cp\u003e3.3 Feature Extraction and Selection\u003c/p\u003e\n\u003cp\u003eBoth traditional image processing features and CNN-learned features were utilized:\u003c/p\u003e\n\u003cp\u003eTraditional features:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eColor features: Color histograms, color moments, and color coherence vectors\u003c/p\u003e\n\u003cp\u003eTexture features: Gray Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LBP)\u003c/p\u003e\n\u003cp\u003eShape features: Hu moments, Fourier descriptors\u003c/p\u003e\n\u003cp\u003eCNN-based feature extraction: A custom CNN architecture was designed to automatically learn hierarchical features from the preprocessed images.\u003c/p\u003e\n\u003cp\u003eFeature selection was performed using a combination of Principal Component Analysis (PCA) for dimensional reduction of traditional features and attention mechanisms within the CNN to focus on disease-relevant image regions.\u003c/p\u003e\n\u003cp\u003e3.4 CNN Architecture and Model Development\u003c/p\u003e\n\u003cp\u003eWe developed a custom CNN architecture optimized for plant disease classification:\u003c/p\u003e\n\u003cp\u003eAdditionally, we incorporated:\u003c/p\u003e\n\u003cp\u003eAttention mechanism: A spatial attention module was introduced after the third convolutional block to help the network focus on disease-specific regions.\u003c/p\u003e\n\u003cp\u003eResidual connections: Skip connections were added to facilitate gradient flow during training and enable deeper network training.\u003c/p\u003e\n\u003cp\u003eTransfer learning: For comparison, we fine-tuned pre-trained models (ResNet-50, MobileNetV2, and EfficientNet-B0) on our dataset.\u003c/p\u003e\n\u003cp\u003eThe model was trained using:\u003c/p\u003e\n\u003cp\u003eLoss function: Categorical cross-entropy\u003c/p\u003e\n\u003cp\u003eOptimizer: Adam with learning rate of 0.0001\u003c/p\u003e\n\u003cp\u003eBatch size: 32\u003c/p\u003e\n\u003cp\u003eEarly stopping: Patience of 15 epochs monitoring validation loss\u003c/p\u003e\n\u003cp\u003eLearning rate reduction: By factor of 0.2 when validation loss plateaued\u003c/p\u003e\n\u003cp\u003e3.5 Model Deployment\u003c/p\u003e\n\u003cp\u003eThe trained model was optimized for mobile deployment through:\u003c/p\u003e\n\u003cp\u003eModel quantization: Weights were quantized to 8-bit integers, reducing model size by approximately 75%.\u003c/p\u003e\n\u003cp\u003ePruning: Non-essential connections were pruned, further reducing computational requirements.\u003c/p\u003e\n\u003cp\u003eTensorFlow Lite conversion: The optimized model was converted to TensorFlow Lite format for efficient mobile execution.\u003c/p\u003e\n\u003cp\u003eA cross-platform mobile application was developed with the following features:\u003c/p\u003e\n\u003cp\u003eReal-time image capture and analysis\u003c/p\u003e\n\u003cp\u003eOffline operation capability\u003c/p\u003e\n\u003cp\u003eDisease information and treatment recommendations\u003c/p\u003e\n\u003cp\u003eHistorical tracking of detections\u003c/p\u003e\n\u003cp\u003eIntegration with agricultural extension services\u003c/p\u003e"},{"header":"4. RESULTS AND DISCUSSIONS","content":"\u003cp\u003eIn this section all the results and the discussions should be made.\u003c/p\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Performance Evaluation\u003c/h2\u003e\u003cp\u003eThe performance of our system was evaluated using standard metrics:\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003cp\u003eProportion of correctly classified instances\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003cp\u003eTrue positives divided by predicted positives\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003cp\u003eTrue positives divided by actual positives\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eF1-score\u003c/strong\u003e\u003cp\u003eHarmonic mean of precision and recall\u003c/p\u003e\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the overall performance of different models tested:\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance comparison of different models\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrecision (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRecall (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1-score (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCNN Proposed\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e97.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e97.65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e97.72\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e97.68\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eResNet-50\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e96.54\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e96.38\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e96.41\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e96.39\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMobileNetV2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e95.82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e95.66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e95.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e95.68\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEfficientNet-B0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e96.93\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e96.81\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e96.75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e96.78\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTraditional ML (SVM with HOG features)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e84.21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e83.95\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e84.17\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e84.06\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eOur custom CNN architecture achieved the highest performance across all metrics, with an overall accuracy of 97.89%. The confusion matrix analysis revealed that most misclassifications occurred between visually similar diseases affecting the same plant species.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Impact of Preprocessing Steps\u003c/h2\u003e\u003cp\u003eTo evaluate the contribution of different preprocessing steps, ablation studies were conducted. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the impact of each preprocessing component:\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eImpact of preprocessing steps on model accuracy\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePreprocessing Configuration\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eComplete pipeline\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e97.89\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWithout background removal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e95.37\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWithout illumination normalization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e96.12\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWithout data augmentation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e94.86\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWithout color space transformation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e97.04\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBasic preprocessing only (resize\u0026thinsp;+\u0026thinsp;normalize)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e92.55\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe results demonstrate that each preprocessing component contributes to the overall performance, with data augmentation and background removal showing the most significant impact.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e4.3 Performance across Different Environmental Conditions\u003c/h2\u003e\u003cp\u003eThe system's robustness was tested across various environmental conditions. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e summarizes the accuracy across different imaging scenarios:\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAccuracy under different environmental conditions\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCondition\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eControlled environment (lab setting)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e99.23\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNatural outdoor lighting\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e96.82\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLow light conditions\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e95.41\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVarying backgrounds\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e96.25\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDifferent distances\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e97.14\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEarly-stage disease symptoms\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e93.87\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eWhile performance remained high across most conditions, early-stage disease detection presented the greatest challenge, as expected. This highlights an area for future improvement.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e4.4 Computational Efficiency and Mobile Performance\u003c/h2\u003e\u003cp\u003eThe optimized model demonstrated excellent performance on mobile devices:\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eMobile performance metrics\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMetric\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eValue\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel size\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e8.7 MB\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAverage inference time (mid-range smartphone)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e312 ms\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMemory usage\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e145 MB\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBattery consumption (per 100 inferences)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e~\u0026thinsp;1%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThese results confirm the system's suitability for deployment on resource-constrained devices, making it accessible to farmers with basic smartphone hardware.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e4.5 Comparison with Existing Systems\u003c/h2\u003e\u003cp\u003eOur system was compared with other published plant disease detection approaches:\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eComparison with existing approaches\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eApproach\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDataset Size\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNumber of Classes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMobile Compatible\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eProposed\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e38,000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e40\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e97.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMohanty et al. (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e54,306\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e38\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e99.35*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFerentinos (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2018\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e87,848\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e99.53*\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRamcharan et al. (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2017\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2,756\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e93.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eToo et al. (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2019\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e54,306\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e38\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e98.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e*Accuracy on controlled environment images only. Performance drops significantly (15\u0026ndash;30%) on real-field images.\u003c/p\u003e\u003cp\u003eWhile some approaches report marginally higher accuracy on controlled datasets, our system maintains high performance across real-world conditions while being optimized for mobile deployment.\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Field Application and User Study","content":"\u003cp\u003eA pilot study was conducted with 50 farmers across different agricultural regions to evaluate the practical utility of the mobile application. Participants used the application for three months during a growing season. Key findings include:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e94% of participants reported the application was \"easy\" or \"very easy\" to use\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eDisease identification by the app matched expert diagnosis in 92% of cases\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e87% of farmers reported earlier disease detection compared to their usual practices\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e76% reported reduced pesticide usage due to more targeted and timely interventions\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eAverage estimated crop loss reduction was 23% compared to previous seasons\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eQualitative feedback highlighted the value of offline functionality and the integrated treatment recommendations.\u003c/p\u003e"},{"header":"5. CONCLUSION","content":"\u003cp\u003eThis paper presented a comprehensive AI-powered system for detecting and classifying plant diseases using image processing and deep learning techniques. Our approach combines robust preprocessing with a custom CNN architecture optimized for mobile deployment. The system achieves 97.89% classification accuracy across a diverse dataset of 38,000 images spanning 14 crop species and 26 diseases.\u003c/p\u003e\u003cp\u003eThe field application study demonstrates the practical utility of the system, with farmers reporting earlier disease detection, reduced pesticide usage, and decreased crop losses. The mobile implementation makes advanced disease diagnostic technology accessible to farmers in diverse agricultural settings, including resource-limited regions.\u003c/p\u003e\u003cp\u003eFuture work will focus on:\u003c/p\u003e\u003cp\u003eExpanding the disease database to include more crop species and disease categories\u003c/p\u003e\u003cp\u003eImproving early-stage disease detection through temporal analysis of plant development\u003c/p\u003e\u003cp\u003eIncorporating environmental data (temperature, humidity, soil conditions) to enhance diagnostic accuracy\u003c/p\u003e\u003cp\u003eDeveloping region-specific models that account for local disease prevalence and manifestation\u003c/p\u003e\u003cp\u003eImplementing cloud synchronization for continuous model improvement through federated learning\u003c/p\u003e\u003cp\u003eThe system presented in this paper contributes to sustainable agricultural practices by providing a cost-effective tool for early plant disease detection, potentially reducing crop losses and environmental impact of excessive pesticide use.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBarbedo, J.G.A. (2013). Digital image processing techniques for detecting, quantifying and classifying plant diseases. SpringerPlus, 2(1), 660.\u003c/li\u003e\n\u003cli\u003eFerentinos, K.P. (2018). Deep learning models for plant disease detection and diagnosis. Computers and Electronics in Agriculture, 145, 311-318.\u003c/li\u003e\n\u003cli\u003eFuentes, A., Yoon, S., Kim, S.C., \u0026amp; Park, D.S. (2017). A robust deep-learning-based detector for real-time tomato plant diseases and pests recognition. Sensors, 17(9), 2022.\u003c/li\u003e\n\u003cli\u003eHughes, D.P., \u0026amp; Salath\u0026eacute;, M. (2015). An open access repository of images on plant health to enable the development of mobile disease diagnostics. arXiv preprint arXiv:1511.08060.\u003c/li\u003e\n\u003cli\u003eJohannes, A., Picon, A., Alvarez-Gila, A., Echazarra, J., Rodriguez-Vaamonde, S., Navajas, A.D., \u0026amp; Ortiz-Barredo, A. (2017). Automatic plant disease diagnosis using mobile capture devices, applied on a wheat use case. Computers and Electronics in Agriculture, 138, 200-209.\u003c/li\u003e\n\u003cli\u003eMohanty, S.P., Hughes, D.P., \u0026amp; Salath\u0026eacute;, M. (2016). Using deep learning for image-based plant disease detection. Frontiers in Plant Science, 7, 1419.\u003c/li\u003e\n\u003cli\u003eRamcharan, A., Baranowski, K., McCloskey, P., Ahmed, B., Legg, J., \u0026amp; Hughes, D.P. (2017). Deep learning for image-based cassava disease detection. Frontiers in Plant Science, 8, 1852.\u003c/li\u003e\n\u003cli\u003eRother, C., Kolmogorov, V., \u0026amp; Blake, A. (2004). GrabCut: Interactive foreground extraction using iterated graph cuts. ACM Transactions on Graphics, 23(3), 309-314.\u003c/li\u003e\n\u003cli\u003eSladojevic, S., Arsenovic, M., Anderla, A., Culibrk, D., \u0026amp; Stefanovic, D. (2016). Deep neural networks based recognition of plant diseases by leaf image classification. Computational Intelligence and Neuroscience, 2016, 3289801.\u003c/li\u003e\n\u003cli\u003eStrange, R.N., \u0026amp; Scott, P.R. (2005). Plant disease: A threat to global food security. Annual Review of Phytopathology, 43, 83-116.\u003c/li\u003e\n\u003cli\u003eToo, E.C., Yujian, L., Njuki, S., \u0026amp; Yingchun, L. (2019). A comparative study of fine-tuning deep learning models for plant disease identification. Computers and Electronics in Agriculture, 161, 272-279.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Malla Reddy Deemed to be University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"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, Convolutional neural networks, Image processing, Deep learning, Precision agriculture, Mobile applications","lastPublishedDoi":"10.21203/rs.3.rs-7218586/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7218586/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis paper presents a novel approach to automated plant disease detection and classification using advanced image processing and deep learning techniques. Early detection of plant diseases is crucial for sustainable agricultural practices and food security. Our proposed system leverages convolutional neural networks (CNNs) to analyze leaf images and accurately identify various plant diseases across multiple crop species. The methodology includes image preprocessing, segmentation, feature extraction, and classification using a custom CNN architecture. The system was trained and validated on a diverse dataset containing 38,000 images spanning 14 crop species and 26 diseases. Experimental results demonstrate 97.89% classification accuracy, outperforming existing methods. The system is implemented as a lightweight mobile application allowing farmers to diagnose plant diseases in real-time using only a smartphone camera, potentially reducing crop losses and pesticide usage through early intervention. This research contributes to precision agriculture by providing an accessible, cost-effective tool for disease management in both developed and developing agricultural contexts.\u003c/p\u003e","manuscriptTitle":"AI-Powered System for Detecting and Classifying Plant Diseases using Image Processing Techniques","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-30 07:53:34","doi":"10.21203/rs.3.rs-7218586/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":"17a42ed8-d7ea-4aa2-9205-cc3bacaaeed7","owner":[],"postedDate":"July 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":52153331,"name":"Computer Architecture and Engineering"}],"tags":[],"updatedAt":"2025-08-20T05:33:46+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-30 07:53:34","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7218586","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7218586","identity":"rs-7218586","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}