An Improved YOLO11 Method Based on D-LKA for Photovoltaic Panel Defect Detection

Photovoltaic (PV) panel defect detection is important to enhance the reliability of PV power generation systems. Traditional PV panel defect detection methods suffer from limitations such as difficulty in large-scale application, sensitivity to dust, and low detection accuracy for small targets. To address these issues, this paper proposes an improved YOLO11 method based on deformable large kernel attention (D-LKA), named YOLO-LKA. The proposed YOLO-LKA uses YOLO11 as the baseline, leveraging its advantages of detection accuracy, speed, and multi-scale feature fusion. The D-LKA module is introduced into YOLO11 to enhance the model's performance with a flexible attention mechanism and the advantages of deformable convolution. The accuracy of small target detection and the robustness of the model are improved. Experiments validate the effectiveness and superiority of the proposed model, showing a significant improvement in small target detection accuracy.