Crop Disease Detection via an Improved Residual Network

Abstract

Deep learning has shown substantial potential across many domains, and crop disease detection is no exception. This study investigates the application of deep learning to crop disease detection, systematically evaluating the performance of convolutional neural networks of different depths. Building on the classical ResNet-50 as the base model, we introduce a Squeeze-and-Excitation (SE) attention mechanism to enhance feature extraction. The results indicate that increasing network depth enables the extraction of more hierarchical and discriminative features, which is beneficial for distinguishing visually similar crop disease categories. Ablation experiments further show that adding the Squeeze-and-Excitation (SE) module to ResNet-50 increases validation accuracy from 80.7% to 85.2%, indicating that channel-wise attention helps emphasize disease-relevant feature channels while suppressing redundant information. Overall, deep learning provides a reliable technical foundation for crop disease detection and yields substantial performance gains.