<![CDATA[Rice leaf diseases such as blast, brown spot, bacterial blight, and tungro pose a serious threat to agricultural productivity, with potential yield losses exceeding 50% under epidemic conditions. Therefore, rapid and accurate early detection is essential to support sustainable food security and precision agriculture. This study proposes a rice leaf disease classification system based on the MobileNetV2 architecture optimized using Particle Swarm Optimization (PSO). A dataset consisting of four rice leaf disease classes, collected from public repositories and sources representing real field conditions, is used to evaluate the proposed approach. Transfer learning is employed to improve training efficiency, while PSO is applied to optimize key hyperparameters to enhance model stability and convergence.Experimental results show that MobileNetV2 optimized with PSO consistently achieves superior classification performance and improved training stability compared to the standard MobileNetV2 baseline. The baseline MobileNetV2 achieves 92% accuracy, with the highest F1-score of 0.99 on the tungro class and the lowest performance of 0.88 on the blast class. In contrast, MobileNetV2–PSO demonstrates a significant improvement, reaching 99% accuracy, with F1-scores of 0.98 for bacterial blight, 0.98 for blast, 0.99 for brown spot, and 1.00 for tungro. The largest improvement occurs in the blast class, with a 10-point increase in F1-score, indicating that PSO optimization provides greater sensitivity to complex disease patterns that were previously difficult to classify.These findings indicate that the proposed framework provides an accurate and lightweight solution with strong potential for deployment on mobile and resource-constrained platforms to support intelligent rice disease diagnosis.]]>
