Improved Yolov5 Algorithm for Surface Defect Detection of Solar Cell

Solar cells are a significant component of photovoltaic power systems, and the quality of the cells directly affects the efficiency of power generation. Aiming to address the difficulty of defect detection caused by complex solar cell backgrounds and diverse defects in the mixed scenario of monocrystalline cells and polycrystalline cells, this paper proposes an improved YOLOv5 algorithm. Firstly, the global attention mechanism (GAM) is incorporated into the backbone network of yolov5 to enhance the feature extraction ability for electroluminescence (EL) images; then, the feature fusion ability of the model is improved due to the introduction of the approach named adaptively spatial feature fusion (ASFF). The experimental results show that the proposed method trained on the solar cell EL image dataset achieves 75.9% mAP, which is 3.5% higher than the original yolov5. In addition, compared with mainstream object detection models, it is further verified that our model has outstanding performance in the solar cell surface defect detection task.