AE-YOLO: Research and Application of the YOLOv11-Based Lightweight Improved Model in Photovoltaic Panel Surface Intelligent Defect Detection

With the rapid development of renewable energy, surface defect detection of photovoltaic panels has become an important link in improving photoelectric conversion efficiency and ensuring safety. However, there are various types of surface defects on photovoltaic panels with complex backgrounds, and traditional detection methods face challenges such as low efficiency and insufficient accuracy. This article proposes a lightweight improved model AE-YOLO (YOLOv11+Adown +ECA) based on YOLOv11, which improves detection performance and efficiency by introducing a lightweight dynamic down-sampling module (Adown) and an Efficient Channel Attention (ECA). The Adown module reduces the complexity of computational and parameters through steps such as average pooling preprocessing, channel dimension segmentation, branch feature processing, and feature fusion. The ECA mechanism enhances the model's response to defect sensitive feature channels and improves its ability to discriminate low contrast small defects through adaptive average pooling, one-dimensional convolution, and sigmoid activation. The experimental results indicate that the AE-YOLO model performs well on the PVEL-AD dataset. mAP@0.5 reached 90.3%, the parameter count decreased by 18.7%, the computational load decreased by 19%, and the inference speed reached 259.56 FPS. The ablation experiment further validated the complementarity between Adown and ECA modules, providing an innovative solution for real-time and accurate defect detection of photovoltaic panels in industrial scenarios.