Residential Solar Panel Object Detection Based on Multi-combination Data Augmentation and YOLOv5

The photovoltaic (PV) installation capacity is important for the planning and operation of the grid. Image recognition has been a new method to detect PV installation capacity. This paper proposes a residential solar panel (RSP) object detection model based on multi-combination data augmentation and YOLOv5. The RSP dataset is established through public satellite images. The multi-combination data augmentation method is proposed to deal with the small sample problem and the YOLOv5s is utilized as the detection algorithm. Several ablation experiments with conventional data augmentation methods are conducted. Compared to the standard yolov5-based benchmark test, our method can increase the overall average precision by 4%.