A YOLO V11 classification model for tomato leaf diseases and pests based on weighted convolution and gating mechanism

Abstract

Aiming at the problems of insufficient modeling of local fine-grained features and strong interference from redundant background information in tomato leaf disease and pest recognition, this paper proposes an improved yolov11 classification model based on weighted convolution and gating mechanism. The proposed method embeds a weighted convolution module and a C2PSA_CGLU module into the backbone of the yolov11 classification network. The weighted convolution module adaptively enhances discriminative channel features, while the C2PSA_CGLU module dynamically suppresses redundant information through a gating mechanism. Ablation experiments and comparison experiments are conducted on a public tomato leaf disease dataset containing ten categories. Experimental results show that the proposed model achieves a top-1 accuracy of 0.9958, which is 0.64% higher than the baseline yolov11 model. Meanwhile, the computational complexity is reduced to 1.80 GFLOPs and the Fitness value reaches 0.9979. The proposed model maintains high recognition accuracy while ensuring lightweight characteristics and stable convergence, providing an effective solution for refined recognition of agricultural diseases and pests in engineering applications.