Research on Fault Object Detection Method for Photovoltaic Panel UAV Inspection Based on YOLOv5

Photovoltaic panels are the core equipment of photovoltaic power plants and require regular inspections. To improve inspection efficiency, unmanned aerial vehicles are currently mainly used to take infrared photos of photovoltaic panels for inspection. Therefore, how to quickly and accurately detect fault objects from infrared photos of photovoltaic panels has become a hot research topic. The two-stage detection method generally has drawbacks such as slow detection speed and difficulty in detecting video data. This paper adopts the one-stage object detection method YOLOv5 for detection, and preprocesses the original image data according to the characteristics of the problem, improving the model. Firstly, to address the issues of low resolution of infrared images of photovoltaic panels and unclear boundaries between fault and normal areas, grayscale transformation was performed on the original image to improve the contrast of the image area; Secondly, to address the issue of relatively low proportion of fault object data in the entire image, Mosaic-9 was used to expand and enhance the original image data; Finally, the performance of the model was improved by introducing the Squeeze and Exception (SE) attention mechanism module into the model. The experimental results show that the method proposed in this paper can detect faulty objects in real-time in the infrared images of photovoltaic panels captured by drones during inspection. Compared with the original YOLOv5 method, the improved YOLOv5 method has improved Precision, Recall, mAP@0.5 and mAP@0.5:0.95 by about 20 percentage points.