An approach to PV fault defect detection based on computer vision
Zeming Wang, Yuqi Geng, Zeyu Wu · 2023
Solar panels are susceptible to defects such as hot patches and cracks due to environmental and human factors, which can directly affect energy management and power generation quality if not maintained in a timely manner. Computer vision technology provides a viable tool for PV fault defect detection by inputting defective images to train a model that learns the defective features and thus detects PV fault defects. This paper addresses the problem of uneven speed and accuracy of PV defect detection, and proposes an improved method based on YOLOv5, which uses the C2f module in YOLOv8 to replace the C3 module in the backbone network, enriching the gradient flow information while ensuring light weight, and embedding the SimAM attention mechanism in the Neck network to improve the detection capability of the network for low resolution as well as small targets The final experimental results show that our proposed network is not only lightweight but also rich in gradient flow information. The final experimental results show that our proposed method improves the accuracy by 1.2%, reduces the number of parameters and computation by 25.3% and 16.5% respectively, and has the same inference speed than the original algorithm. Thus, our proposed method achieves a balance between speed and accuracy.