He, Shuyi (Beijing jiaotong University), Cao, Yuan (Beijing Jiaotong University), Wang, Feng (Beijing Jiaotong University), Sun, Yongkui (Beijing Jiaotong University), Su, Shuai (Beijing Jiaotong university), Yang, Weifeng (the CRRC Zhuzhou Institute Co, Ltd), wang, Wenkun (CRRC Zhuzhou Institute Co, Ltd)

Patchcore-SAM : Rail Defect Unsupervised Detection and Segmentation

Scheduled for presentation during the Regular Session "Rail Traffic Management I" (ThAT7), Thursday, September 26, 2024, 11:50−12:10, Salon 15

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 3, 2024

Abstract

The heavy-haul railways plays a crucial role in facilitating the transportation of large-scale goods, yet the challenging operational conditions significantly increase the occurrence of rail defects. Despite the growing importance of computer vision and artificial intelligence in detecting rail defects, current methods face issues of low accuracy and efficiency. The success of the Segment Anything Model (SAM) provides a promising solution to these problems. In this study, a novel two-stage rail defect evaluation methodology called Patchcore-SAM was proposed. Firstly, Patchcore enables the unsupervised detection and localization of defects. Subsequently, the detected anomalies are transformed into anomaly images and prompts, which serve as inputs for the fine-tuned SAM to achieve precise rail defect segmentation. The experimental results demonstrate the superior performance of this method, providing robust support for ensuring the safe operation of heavy-haul railways.