ResNet-Based Multiscale Feature Extraction and Attention Mechanism Integration for Hazelnut Disease Classification

Abstract

This study aims to develop an artificial intelligence-based classification model for early detection of plant diseases that cause yield loss in hazelnut production in Turkey. Apple mosaic virus, Armillaria root rot, hazelnut bacterial blight, powdery mildew and Ganoderma root rot diseases were analyzed with images obtained from the Black Sea Region. The proposed model is based on ResNet-50 architecture augmented with Squeeze-and-Excitation (SE) blocks, including multi-scale feature extraction and channel-based attention mechanism. Evaluated with 10-fold cross-validation, the model achieved the highest success with 89.53% accuracy and 89.31% F1 score. The results show that AI-based early detection systems have significant potential in disease detection and improving agricultural productivity. © 2025 IEEE.