Oryza YOLOCNN-X: Real-time Rice Disease Detection and Classification on Raspberry Pi

Abstract

Rice is significantly impacted by its disease leading to loss of rice field yield impacting global food security and thereby effective real-time monitoring solutions are required. This proposed study created a deep learning-based framework for rice disease detection. The proposed framework integrates a patch-based disease extraction and classification using deep learning model which can be used for drone-captured imagery. A YOLO-based model for disease patch extraction and texture based-CustomCNN classification for disease identification. The framework is deployed on real-time edge-device i.e raspberry pi. The training of model is done on dataset of image count 5,181 belonging to 9 disease classes captured in Hyderabad and then testing was done on self-captured images from the rice field captured in Palghar, thus demonstrating cross-region validation. The experimental results demonstrate that model training, validation and testing gave an accuracy of 96.67,98.07,94.46 respectively, thus when deployed on edge devices gave 52 ms latency and 18.5 fps.