Based on Transfer Learning a Performance Analysis and Comparative Study of CNN for Accurate Plant Disease Detection

This paper investigates the use of Convolutional Neural Networks (CNNs) for plant disease diagnosis, utilizing the notion of transfer learning to develop a deep CNN network at a low cost. The dataset utilized includes 30,000 pictures from 18 distinct categories. Several architectural models, including ResNet50, ResNet50V2, VGG16, VGG19, EfficientNetB0, MobileNet, MobileNetV2, NASNetMobile, and DenseNet, were created using transfer learning methodologies. The models are trained on the target plant disease dataset while keeping the features learnt from large-scale image datasets like ImageNet. Performance criteria such as accuracy, precision, and recall are used to assess the models' accuracy in detecting various plant diseasesx. Furthermore, the research investigates the effect of dropout on accuracy and other model parameters in the context of transfer learning. Finally, the outcomes of a tailored model with and without batch normalization are compared.