铁道标准设计2025，Vol.69Issue(8)：55-63,9.DOI:10.13238/j.issn.1004-2954.202311290004
无砟轨道关键部位表观病害机器视觉识别方法研究
Research on Machine Vision-Based Identification Methods for Visible Defects in Critical Components of Ballastless Tracks
摘要
Abstract
Ballastless tracks are widely applied in China's high-speed railway system,and the wide and narrow joints are among the key components to ensure the integrity and stability of the track system.Only through rapid and accurate identification of visible defects at critical locations can efficient and scientific maintenance be supported.To better balance detection accuracy,speed,and model size in scenarios with poor lighting conditions and high similarity between defect characteristics and background,this study proposed YG-E-Mv2Net-b,a machine vision-based identification method for visible defects,based on improved YOLOX.Firstly,an Efficient Channel Attention(ECA)module was added between the backbone and neck networks to suppress background information similar to concrete and more accurately extract multi-scale(large,medium,and small)defect target features.Secondly,the position loss function was optimized from IoU to GIoU to address the issue of inaccurately reflecting the overlap between the predicted and ground truth boxes,thereby improving the detection accuracy of defect locations.Finally,the backbone network was replaced with the lightweight MobileNetV2,significantly reducing model parameters and computational cost while enhancing both accuracy and speed.The experimental results showed that the method achieved a recognition accuracy of 97.6％.Compared to YOLOX,the identification speed increased by 3 Img/s,computational cost decreased by 9.57 GFLOPs,and the number of parameters was reduced by 5.8 M,effectively balancing accuracy,speed,and model size.关键词
无砟轨道/宽窄接缝/病害识别/YOLOX/注意力机制/MobileNetV2Key words
ballastless track/wide and narrow joints/defect identification/YOLOX/attention mechanism/MobileNetV2分类
交通工程引用本文复制引用
李培刚,张瑞心,李文举,孙宏杰,王璐,刘泽轩..无砟轨道关键部位表观病害机器视觉识别方法研究[J].铁道标准设计,2025,69(8):55-63,9.基金项目
"一带一路"中老铁路工程国际联合实验室科研项目(21210750300) （21210750300）
中国国家铁路集团有限公司科技研究开发计划重大课题(K2020G031) （K2020G031）
