Performance Analysis of Lightweight CNN models on fungi crops disease image dataset of maize, sorghum and rice.

Crop health is a critical determinant of food security, especially in developing regions where staple crops like maize, sorghum, and rice serve as primary nutrition sources. However, fungal diseases continue to threaten these crops, leading to significant yield losses. Traditional disease detection methods are largely manual and prone to inaccuracies, prompting the need for automated, scalable solutions. This study explores the application of lightweight Convolutional Neural Networks (CNNs) for accurate and efficient classification of fungal diseases in crop leaves. Five pre-trained lightweight CNN architectures—VGG16, MobileNetV3, DenseNet121, NasNetMobile, and EfficientNetB2 were evaluated using a curated dataset of diseased and healthy leaf images. The models were assessed based on classification accuracy, computational efficiency, model size, and suitability for deployment in low-resource environments. Experimental results showed that EfficientNetB2 consistently outperformed other models in accuracy across all three crops, with MobileNetV3 providing a viable alternative for edge-device deployment due to its minimal memory footprint and low inference time. This research demonstrates the feasibility of deploying lightweight CNNs for early disease detection, supporting the advancement of precision agriculture in resource-constrained settings.