<![CDATA[Rice is a critical commodity for national food security; however, its productivity is frequently reduced due to leaf diseases and pests. Conventional identification methods that rely on visual observation are often inefficient and prone to subjectivity, particularly given the complex and variable nature of symptoms. This study to evaluate and compare the performance of several lightweight CNN architectures in accurately and efficiently detecting rice leaf diseases and pests on resource constrained devices. This study compares four CNN lightweight architectures: MobileNetV2, EfficientNetV2-B3, NasNetMobile, and a custom CNN Lightweight Architecture, all using a 13-class dataset that underwent preprocessing, augmentation, and data balancing. The models were trained for 100 epochs using the Adam optimizer. Experimental results show that EfficientNetV2B3 achieved the best performance, with 97% accuracy, precision, recall, and F1-score, followed by MobileNetV2 and NasNetMobile, which achieved 94% accuracy. The Custom CNN lightweight model produced 91% accuracy with a model size of only 0.53 MB. Overall, this study provides recommendations for developing accurate and efficient lightweight CNN models to support rice disease and pest detection on mobile devices, IoT systems, and edge computing platforms.]]>
Copyrights © 2025
