Comparative Study of Lightweight CNN Architectures for Maize Leaf Disease Detection

The study evaluates different compact Convolutional Neural Networks (CNNs) used to detect maize leaf diseases because they serve vital functions in precision agriculture. Testing involved evaluating the performance of five various models including VGG19, ResNet50, MobileNetV3, Custom MobileNetV3 and Custom InceptionV3 for the detection of four maize leaf disease types namely Blight, Common Rust, Gray Leaf Spot and Healthy. The analysis demonstrates that Custom MobileNetV3 surpasses all competing models through its 97.63% accuracy and 96.68% precision rating as well as 97.96% recall value. The model showed complete ability to detect Gray Leaf Spot disease which indicated its exceptional efficiency for spotting this condition. The ResNet50 model displayed good performance by effectively detecting Common Rust together with Healthy maize leaves. The recall level was lower for Gray Leaf Spot based on the results observed by the model. The detection results for Gray Leaf Spot demonstrate that Custom MobileNetV3 surpassed both VGG19 and MobileNetV3. The poorest performance of Custom InceptionV3 resulted from Wheat Blight and Gray Leaf Spot being confused with one another. Custom MobileNetV3 stands out as the best model for maize disease detection since it delivers reliable performance while maximizing efficiency thus making it appropriate for limited resource scenarios. The study contributes useful information which helps optimize machine learning models applicable for agricultural field usage.