APPLICATION OF DEEP LEARNING ARCHITECTURES TO IDENTIFICATION OF TREE DISEASES FROM IMAGES

Verezubova N., Yakovleva O., Chekulaev A.

This paper focuses on automating the diagnosis of woody plant pathologies using deep learning methods. The relevance of the study lies in the need for timely detection of fungal infections to minimize environmental and economic damage. A dataset of leaf images was used, including healthy specimens and those affected by Uromyces appendiculatus (rust) and Mycosphaerella angulata (angular leaf spot). The methodology is based on a deep neural network with a contrastive learning strategy (SimCLR) and subsequent supervised retraining. The Grad-CAM attention visualization algorithm was used to verify decisions and ensure the interpretability of the model. The results confirm the effectiveness of the proposed approach for accurate disease identification and its applicability to intelligent plant monitoring systems.

Keywords:neural networks, computer vision, plant pathology, contrastive learning.