Semi-Supervised Rice Leaf Disease and Pest Classification Using IoT Field Images

Abstract

Rice is a vital crop for global food security, but its productivity is frequently threatened by diseases and pests, necessitating rapid and accurate detection. This study presents a semi-supervised deep learning approach to automate the classification of ten categories of rice leaf conditions, including seven diseases, two pests, and a healthy class. Field images were acquired using a custom solar-powered ESP32-CAM- based Internet of Things (IoT) edge node. To leverage unlabeled data and minimize manual annotation costs, the dataset was structured with 744 expert-labeled images and 10,249 unlabeled images. An initial Convolutional Neural Network (CNN) model was trained using the 744 expert-labeled images to generate pseudo-labels for the 10,249 unlabeled images. The combined labeled and pseudo-labeled dataset was subsequently partitioned into 80% training, 15% testing, and 5% validation sets to train the final transfer learning-based CNN and Support Vector Machine (SVM) models. Under the evaluated configurations, the CNN achieved higher performance than the SVM baseline, with an accuracy of 82.9%, precision of 83.0%, recall of 82.9%, F1-score of 0.829, and Matthews Correlation Coefficient (MCC) of 0.798, whereas the evaluated SVM configuration achieved an accuracy of 23.7%. These findings demonstrate the efficacy of combining low-cost IoT edge acquisition with semi-supervised deep learning for scalable, real-time crop disease monitoring in precision agriculture.