Intelligent Recognition Method of Substation Pointer Instrument Based on Deep Learning
Fanglong Niu, Shanxiang Wang, Bo Yu, Xuesong Huo · 2022
Instrument patrol inspection of substation is an important part of intelligent operation and maintenance of substation. To replace manual patrol inspection to realize automatic patrol inspection of substation instruments, this paper designs a strategy combined with deep learning and computer vision technology. Firstly, the instrument area is detected and located by using object detection algorithm YOLOv4, and then the dashboard area is segmented by using U-Net network based on the detection results; After that, ellipse fitting and perspective transformation are performed on the segmentation results; Then, the introduction of attention mechanism in U-Net is analyzed and the U-Net network is used to segment the scale area and pointer area. Finally, the computer vision technology is used for fitting processing and combined with the prior information get by YOLOv4, the instrument reading is obtained. The results shows that recognition is basically similar from this strategy to manual reading, and the “machine generation” can be realized.