Defect Detection of Photovoltaic Panel Electroluminescence Image Based on YOLOv7-SE-DS-NWD Algorithm

Solar energy serves as a representative of renewable energy sources, and defects in photovoltaic panels are crucial factors affecting the solar power generation system. To enhance the power generation efficiency of solar energy, a defect detection algorithm for electroluminescence images of photovoltaic panels based on YOLOv7-SE-DS-NWD is proposed. First, in YOLOv7 (you only look once version 7), extended efficient layer aggregation networks (extended efficient layer aggregation networks, ELAN added the squeeze-and-excitation (SE) attention mechanism to make the model more focused on useful channel information. Then, the convolution module in the detection head of YOLOv7 is replaced with Distribution Shifting Convolution (DSConv). In the detection head, the regression loss function replaces IoU (intersection over union) with the Normalized Gaussian Wasserstein Distance (NWD) to enhance small target detection capability. According to experimental results, the improved YOLOv7 algorithm exhibits a 1.4% increase in detection accuracy compared to the original YOLOv7. This enhancement addresses missed and false detections of defects in electroluminescence images of photovoltaic panels, optimizing the overall defect detection performance for photovoltaic panel applications in power plants.