Improved Solar Photovoltaic Panel Defect Detection Technology Based on YOLOv5

Nowadays, the photovoltaic industry has developed significantly. Solar photovoltaic panel defect detection is an important part of solar photovoltaic panel quality inspection. Aiming at the problems of chaotic distribution of defect targets on photovoltaic panels, large scale span and blurred features, this paper improves the network structure based on the YOLOv5 model, which can better cope with the defect detection under various conditions. This paper mainly optimizes the following three aspects. Firstly, for the defect targets that are not of the right scale, the SE attention module and GhostBottleneck are introduced on the basis of the YOLOv5 model, which accelerates the extraction of useless features and enhances the precision of small object perception by capturing the features of defects of different scales and fusing semantic features of different depths. Secondly, Ghostconv is introduced to increase the receptive scope of the feature map, which makes the extracted feature discrimination ability stronger, and effectively enhances the defect detection ability of the model. Finally, by utilizing the ELU activation function instead of the Leaky ReLU activation function, the problem of gradient explosion and gradient disappearance is solved, and the model convergence speed is accelerated. Experimental results demonstrate that the improved YOLOv5 model can effectively detect the defects of photovoltaic panels, and the mAP reaches 92.4%, which is 16.2% higher than the original algorithm.