Design and Analysis of Plant Disease Identification Using Deep Learning

Abstract

Smart Farming leverages modern Information and Communication Technologies (ICT) to enhance agricultural efficiency and sustainability. Plant diseases put a major disruption to the global food chain by reducing crop quality and yield while also introducing potential health risks. Classical methods that were used earlier for disease diagnosis rely on expert visual inspection, which is labour-intensive, costly, and impractical for large-scale farming. By the time symptoms are visible, significant crop damage (10-20%) may have already occurred. Therefore, developing automated, early-stage disease detection systems is critical for mitigating yield loss and ensuring timely intervention. newlineThis research focuses on using deep learning methods for automatic detection and categorization of plant diseases, evaluating multiple pre-trained deep convolutional neural networks on datasets comprising pepper, potato, tomato, medicinal plants, and rice plants. The study investigates the performance of various pre-trained models like Visual Geometry Group (VGG16, VGG19), Residual Network (ResNet50), Densely Connected Convolutional Network (DenseNet121), Inception Version 3 (InceptionV3), Extreme Inception (Xception), Neural Architecture Search Network Mobile (NASNetMobile), Inception Residual Network Version 2 (InceptionResNetV2), and Mobile Network (MobileNet) with VGG16 achieving the highest accuracy in top-1 and top-5 evaluations. To improve model robustness and generalization, data augmentation techniques were applied to mitigate overfitting. newline newline