Scholarlink

<p>Efficient safety inspections of railway facilities are crucial for urban rail systems, particularly in densely populated cities. This study introduces an advanced railway inspection system integrating a rail-scanning robot with a deep learning-based damage detection framework. The robot captures images at multiple points on the rail, using an onboard computing device to record the images and their locations. Four images are captured at each inspection point to cover the lanes and railway sleepers. After scanning a designated area, the images are sent to a central server for processing using a cascade deep learning framework. This framework employs dual-object detection and image classification networks. Initially, a rail corrugation classification network analyzed lane images to identify railhead corrugations. Subsequently, all images are examined using a concrete damage detection network and railway member detection network, which can identify six types of railway damage. The effectiveness of the proposed system is evaluated on a 550-m railway, achieving an F1-score of 95.5%. The framework proposed in this study significantly reduces the total testing time compared to traditional human-led inspections. The innovative system is expected to enhance the maintenance and safety of urban railway networks by generating faster inspection reports.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Real-time urban railway damage detection-based cascade deep learning networks using automated inspection robot

  • 期刊论文

  • 发表日期：

    2025年6月27日

查看全文
  • Jeongmin Lee,
  • Byunghyun Kim,
  • Dongwoo Kim,
  • Soojin Cho

摘要

Efficient safety inspections of railway facilities are crucial for urban rail systems, particularly in densely populated cities. This study introduces an advanced railway inspection system integrating a rail-scanning robot with a deep learning-based damage detection framework. The robot captures images at multiple points on the rail, using an onboard computing device to record the images and their locations. Four images are captured at each inspection point to cover the lanes and railway sleepers. After scanning a designated area, the images are sent to a central server for processing using a cascade deep learning framework. This framework employs dual-object detection and image classification networks. Initially, a rail corrugation classification network analyzed lane images to identify railhead corrugations. Subsequently, all images are examined using a concrete damage detection network and railway member detection network, which can identify six types of railway damage. The effectiveness of the proposed system is evaluated on a 550-m railway, achieving an F1-score of 95.5%. The framework proposed in this study significantly reduces the total testing time compared to traditional human-led inspections. The innovative system is expected to enhance the maintenance and safety of urban railway networks by generating faster inspection reports.