ADVANCED CLASSIFICATION OF POTATO LEAF DISEASES USING EFFICIENT NET V2-S

Abstract

Potato is one of the crops that are important globally, and it is prone to foliar diseases, including the Early Blight and the Late Blight, which cause massive losses in terms of crop yields. The conventional visual inspection techniques are very time consuming and can be subject to errors in distinct circumstances. The present paper presents a hybrid deep learning model, which consists of a customized branch of CNN and EfficientNetV2-S to provide automatic classification of the potato leaf disease. The extraction of high-level and detailed lesion features can be effectively obtained with the help of the two-step training process in which frozen layers and fine-tuning were involved. The model using 3,000 stratified images and heavy augmentation and class-weighted optimization obtained an accuracy of 99.83 on validation and 99.67 on test with a weighted F1-score, Cohen Kappa and Matthews Correlation Coefficient all above 0.995. The fact that the error margin is close to zero indicates the strength, effectiveness, and possible use of the model in real-time disease diagnostics and precision agriculture