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

Abstract

The potato is an essential crop worldwide, vulnerable to foliar diseases such as Early Blight and Late Blight, which lead to significant yield losses. Traditional visual inspection methods are time-consuming and prone to errors under various conditions. This paper introduces a hybrid deep learning architecture that combines a tailored CNN branch with EfficientNetV2-S to classify potato leaf diseases automatically. The two-step training process, involving frozen layers and fine-tuning, enables effective extraction of high-level and detailed lesion features. Using 3,000 stratified images with extensive augmentation and class-weighted optimization, the model achieved a validation accuracy of 99.83 and a test accuracy of 99.67, with weighted F1-score, Cohen Kappa, and Matthews Correlation Coefficient all above 0.995. The near-zero error rate highlights the model's robustness, efficiency, and potential application in real-time disease detection and precision agriculture.