Intelligent Quality Inspection of Secondary Wiring based on Rotating Object Detection and Pairwise Matching Strategy

Qiejun Dai, Zeyong Mao, Yong Wu, Rui Yao, Jingshi Chen, Lei Chen

Abstract: As a routine work in substations, the quality inspection of secondary wiring is currently done manually by quality inspectors, which is not only time-consuming and laborious, but also boring, making the quality inspectors extremely fatigued and prone to errors. To overcome the above shortcomings of manual quality inspection, we propose an intelligent quality inspection method of substation secondary wiring based on rotating object detection and pairwise matching strategy. Firstly, a data set is constructed, and the image data is collected during on-site inspection of secondary wiring in the substation. The collected images are augmented, and the annotation sample data set is constructed by using annotation tools. Then, the rotating object detection model is trained to locate and recognize the strings on the terminal number plates (TNPs) and wiring caps. However, due to the irregularity of the secondary wiring and inevitable visual distortion in the process of data collection, it is impossible to complete the checking task by using string localization and recognition results directly. Therefore, this paper further designs an effective pairwise matching strategy to judge whether TNPs and wiring caps are correctly paired based on the domain prior knowledge. Finally, the experimental results on the real dataset verify the effectiveness of our method, which can assist manual quality inspection to a certain extent, reduce the work intensity of quality inspectors, and improve work efficiency.