Automatic Detection Strategy of Multi-Scale Catenary Support Device Based on Improved YOLOv7
Dongzhu Jiang, Keyan Liu, Limin Jia, Yong Qin, Yaopeng Jiang, Zhipeng Wang · IFAC-PapersOnLine · 2023
In recent years, UAV(Unmanned Aerial Vehicle) has shown great vitality and potential in automated railway inspection operations. The catenary support device is an important infrastructure on the high-speed railway, which guarantees for the electric power system when the high-speed railway is running. However, the current inspection by UAV cannot automatically obtain multi-scale and standardized image data of the catenary support devices. It presents a serious challenge to the faulty diagnosis of the catenary support devices. Concerning this issue, an automatic multi-scale image capture strategy of catenary support device based on improved YOLOv7 is proposed. Due to the characteristics of less foreground and more background in the image detection of catenary support device, the Focal Loss is introduced. Experimental results show that the convergence speed of the model is improved with high detection accuracy. Furthermore, multi-scale and standardized capture strategies for catenary support device data are proposed. In the deployment experiment, improved YOLOv7 achieves 81.3% mAP, the multi-scale, standardized and normalized image data of catenary support device is captured.