RETRACTED ARTICLE: Anomaly Detection of High-Speed Railway Catenary Damage

Abstract

The Railway Catenary system is an integral and essential component of railway electrification. The working state of the catenary system can directly affect the normal operation of the locomotive. The section insulator (SI) in the catenary system is required for the electrical segregation of two feeds such as in a crossover while deciding the destined track from two adjacent tracks. We propose a deep learning-based approach for abnormal detection of insulator breakage in high-speed railway catenary. Semantic segmentation is an important theory in deep learning-based image segmentation. Aiming at the abnormal detection of insulator breakage in high-speed railway catenary, semantic segmentation technology is used to detect the abnormalities, which can quantitatively describe the area of insulator breakage or damage. Based on the DeepLab V3+ semantic segmentation network, the characteristics of the insulator’s data are retrieved. The Multi Feature DeepLab V3+ network and the Attention DeepLab V3+ networks are proposed to solve the anomaly detection problem of the damaged area from the insulator’s images. The results show that the improved two-folded semantic segmentation networks can better mark the contours of the damaged area of the insulator to safeguard the Catenary system. The outcomes are compared with the original DeepLab V3+ and the MIoU values of the two improved semantic segmentation networks have been improved to a remarkable extent that improves the overall performance of the deep learning-based segmentation of insulator’s images.