RiceGuardAI:An Efficient And LightWeight Deep Learning Framework for Automated Rice Leaf Disease Detection And Agricultural Management

Abstract

Rice plays a vital role in global food supply, serving as a primary dietary source for a significant portion of the world’s population. However, the occurrence of rice leaf diseases considerably affects agricultural productivity, leading to yield losses of nearly 20–30%. Timely and reliable disease identification is therefore essential for improving crop management practices and ensuring food security. Recent developments in deep learning have enabled automated disease diagnosis systems, but deploying such systems in practical agricultural environments requires a balance between prediction accuracy and computational efficiency. This study presents a comparative evaluation of four deep learning architectures, namely Custom Convolutional Neural Network (CNN), MobileNetV2, ResNet50, and InceptionV3, for the classification of four rice leaf conditions: Bacterial Blight, Brown Spot, Leaf Smut, and Healthy leaves. All models were trained and tested on a curated dataset containing 13,760 images using identical experimental settings to ensure fair performance comparison. Among the evaluated architectures, the proposed lightweight Custom CNN achieved the best overall performance with an accuracy of 99.90%, while maintaining only 2.1 million parameters, a compact model size of 9.8 MB, and a real-time inference speed of 12.13 ms. In comparison, MobileNetV2, InceptionV3, and ResNet50 achieved accuracies of 98.93%, 98.79%, and 88.91%, respectively. In addition to model evaluation, a deployment-oriented web-based agricultural platform was developed, integrating real-time disease diagnosis, ensemble-based yield prediction, expert consultation, and community-driven knowledge exchange. The proposed framework highlights the practical applicability of artificial intelligence in precision agriculture and demonstrates its potential for developing scalable and efficient farming solutions suitable for resource-constrained environments.