Photovoltaic Panel Defect Detection Model Enhanced by Multiscale Convolution and Cooperative Attention

Abstract Addressing the issues of multi-type defect recognition and low detection accuracy for small target defects in photovoltaic panel defect detection, an efficient photovoltaic panel defect detection model (TGU-YOLO) based on YOLOv11n is proposed. Firstly, multiple patterns of input features are captured through a triple parallel lightweight convolutional module to enhance the detection performance of the model. Secondly, a GAM-SimAM collaborative attention mechanism is incorporated into the neck network to achieve multi-scale enhancement of photovoltaic panel defect features and improve the feature extraction ability for small target defects. Finally, the Unified-IoU loss function is used to dynamically adjust the weights of prediction boxes, improving the robustness of target defect detection. A public photovoltaic panel defect dataset is employed for systematic evaluation, and ablation experiments, comparative experiments, and robustness tests are conducted. Experimental results show that the accuracy, recall rate, and mAP@0.5 of the TGU-YOLO model are improved by 5.4%, 7.1%, and 3.4% respectively compared to the YOLOv11n model. While maintaining a low computational load, the model achieves high-precision detection of multiple types and small target defects on photovoltaic panels, and its robustness and generalization ability are significantly enhanced.