Recognition and Classification of Cucurbit Diseases Based on ResNet Improved Network Architecture

Automatic plant disease recognition is one of the key steps to realizing the development of smart agriculture, which is especially important for improving crop yield and disease control. In this paper, we take cucurbit crops as the research object and construct a framework for image classification based on the improved ResNet18 network structure, aiming to improve the accuracy and robustness of disease recognition. We introduce the fused attention module (SE-GAM) in the residual block of each stage on the basis of the standard model structure to enhance the model's ability to recognize disease spot features. The results of training and evaluation on a self-constructed dataset containing 12 types of cucurbit leaf disease images show that the introduction of the SE-GAM module effectively improves the classification performance of the model, especially outperforming the original structure in terms of the F1 score, Cohen's kappa, and Matthews' correlation coefficient (MCC), which further verifies the important role of the attentional mechanism in enhancing the model's discriminative power and stability. This further validates the important role of the attention mechanism in improving the discriminative power and stability of the model.