Wheat Leaf Disease Identification Based on a Fusion Model Integrating Attention Mechanisms and Channel-Spatial Reconstruction Convolution

With the continuous growth in food demand and the escalating importance of crop health, the challenges posed by the high costs and low efficiency of traditional wheat leaf disease detection methods have become increasingly acute. In light of the intricacies and low discernibility of the morphological and color characteristics of wheat leaf diseases and pests in natural environments, this study introduces a novel and efficient recognition and classification model, SCC2f-CSPDarknet53-CBAM, offering a precise and rapid solution for identifying diseases and pests, thereby supporting early prevention and scientific management. By incorporating the SCC-C2f module into the base of C2f-CSPDarknet53, this model enhances the efficiency of network feature fusion and the capability to extract critical features. Furthermore, the integration of the ResBlock-CBAM module at the terminal end bolsters the generalization and recognition abilities for targets. The study employs the Wheat Leaf Disease dataset, which is constructed through a blend of online and offline methods, with meticulous data processing, for training purposes. Experimental results reveal that the SCC2f-CSPDarknet53-CBAM model achieves an average recognition accuracy of $99.54 \%$ in the task of wheat leaf disease detection and classification, surpassing mainstream models in recent years, showcasing remarkable accuracy and adaptability.