An Integrated Entropy-Segmentation and Lightweight CNN Framework for Multi-Class Rice Leaf Disease Detection

Rice is a staple food crop and a vital contributor to global food security. However, rice plants are highly susceptible to multiple diseases, including bacterial leaf blight, brown spot, blast, scald, and narrow brown spot, which can drastically reduce yield and quality. Early and accurate detection of these diseases is critical for precision agriculture. In this study, we propose an integrated framework combining entropy-based image segmentation with a lightweight convolutional neural network (MobileNetV2) for multi-class rice leaf disease detection. Entropy-based masks are generated to highlight diseased regions, and the original RGB image is concatenated with the mask to form a 4-channel input for training. The model is trained on a dataset of six rice leaf classes and evaluated using accuracy, precision, recall, and F1-score. The proposed framework achieved an overall accuracy of 91% on the test set, with high per-class performance (F1-scores: bacterial leaf blight 0.99, leaf scald 0.98, healthy 0.94). Furthermore, the model was exported to TensorFlow Lite (TFLite), demonstrating deployment potential on mobile and IoT devices. These results indicate that integrating segmentation and lightweight CNNs provides a scalable and accurate solution for real-time plant disease detection.