Abstract—Leaf disease classification faces significant challenges due to dataset imbalances, particularly within minority classes, leading to decreased model accuracy. This study addresses this problem by introducing EdgeCutMix, a novel image augmentation technique designed to enhance the representation of minority classes. EdgeCutMix integrates edge detection, using the Canny edge detection algorithm, and selective mixing strategies to generate realistic and informative augmented images. The Plant Pathology 2020 dataset, consisting of 3,642 apple leaf images, was used for evaluation. The experimental setup involved oversampling and comparison against existing techniques like MixUp, CutOut, CutMix, and Mosaic, and training on four CNN architectures: MobileNetV2, EfficientNetB7, ResNet50, and DenseNet201. Results showed that EdgeCutMix significantly improved classification accuracy for minority classes, achieving up to 98% accuracy with the EfficientNetB7 model. These findings suggest that EdgeCutMix provides a promising solution for improving model performance in imbalanced datasets, with potential applications in advancing deep learning in agricultural pathology.
 
Keywords—leaf disease classification, image augmentation, minority class, imbalanced dataset, EdgeCutMix


Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.