The Research on a Lightweight Solar Panel Defect Detection Algorithm Based on Improved YOLOv11

Aiming at addressing the problems of insufficient detection accuracy and redundant model parameters in current surface defect detection schemes for solar panels, this work presents the DCS-YOLO object detection algorithm, with the YOLOv11n model serving as its foundation. First, the DilRepCSPELAN feature extraction module is constructed to reduce the complexity of the model, boost the model’s capacity for extracting features of defect targets, and expand the receptive field in an efficient manner. Second, the coordinate attention mechanism is embedded into the HSFPN network to replace the original feature fusion layer of the network, which facilitates the precise capture of the spatial positional information of targets and further lays a foundation for the subsequent detection head to output more accurate target localization and classification features. Finally, the SEAMHead detection head is employed to tackle the issues of feature loss and localization inaccuracy caused by mutual occlusion between certain defects, ultimately resulting in a notable enhancement of detection accuracy. Final results show that the modified model has achieved an improvement in the key evaluation metric mAP@0.5, with the improvement value being 2.67%. Furthermore, the model’s parameter count is reduced by 44.67%, and the computational overhead is lowered by 36.5%. The DCS-YOLO algorithm achieves model lightweighting and offers a viable solution for the application of edge-side detection devices in industrial scenarios.