GCSC-Detector: A Detector for Photovoltaic Cell Defect Based on Deep Learning

Due to the existence of many small and weak defects and strong complex background interference in the electroluminescence(EL) image of photovoltaic cell modules, it leads to a challenging task to detect the defects of photovoltaic cell modules in the industrial field. In this paper, we propose a Global Channel and Spatial Context Module (GCSC), which includes the channel and the spatial self-attention module, to adaptively capture the global rich context information, and establish the relationship between each channel and the relationship between each pixel. Moreover, GCSC achieves the effect of suppressing background noise and enhancing feature representation. Then embed this module into the YOLOv7 model to form our Global Channel and Spatial Context Detector (GCSC-Detector) to improve the detection ability for small and weak defects. The experimental results show that the mAP50 of this method reaches 84.8% on the large-scale photovoltaic EL dataset PVEL-AD and it is superior to other methods.