Automated Paddy Disease Detection: A Deep Learning Approach

Abstract

Paddy production is one of the significant branches of the agricultural sector that is vulnerable to being affected by a range of fungal, bacterial, and viral infections. These diseases can cause significant yield losses and reduce the quality of rice crops if not managed effectively. To address this issue, automation of paddy disease detection plays a vital role in enhancing productivity and promoting sustainable farming. In our study, we presented an innovative deep-learning approach for the automated detection of paddy diseases considering six distinct disease types, including Bacterial Blight, Blast, Brownspot, Leaf scaled, Sheath Blight, and Tungro along with healthy leaves. We utilized a composite dataset of 8526 images collected from Kaggle and Mendeley consisting of seven classes. Two deep learning models: Convolutional Neural Network (CNN) and ResNet-50 were applied in the process. The CNN model achieved an accuracy of 88% on the employed dataset while ResNet-50 exhibited superior performance of 97% accuracy. The model's performance is evaluated using confusion matrices containing precision, recall, accuracy, and F1-score. The pinnacle of the research includes the deployment of the enhanced ResNet-50 model into a mobile application designed using Flutter and FastAPI. This application demonstrates the possibilities of real-time agricultural diagnostics by allowing users to shoot photos of paddy crops and instantly identify probable diseases. It represents a significant advancement in agricultural technology, offering timely data to improve crop health management and maximize harvests.