研究目的: 钢轨表面伤损分割是铁路工务巡检的重要内容。为应对轨面伤损未及时发现导致安全事故发生的难题,解决传统检测方法适用性受限的问题,本文提出一个集成多种深度学习模型的伤损自动化分割算法,为钢轨表面伤损的识别、分析和处理提供一个精准且高效的解决方案。
研究结论: (1)提出了一种集成目标检测和语义分割的深度学习算法,实现了针对钢轨表面伤损特征的高效识别与精准分割;(2)实践结果表明,本文算法与现有的几种深度学习模型相比精度更高,准确度和平均交并比分别达到99.56%和83.89%;(3)本文算法能够精细化地区分钢轨伤损和背景的模糊边界,减少数据冗余,加快分割效率,对多尺度伤损目标的分割准确性高,研究结论可为铁路工务部门的自动化检维提供理论指导。
Abstract
Research purposes: Rail surface damage segmentation is an important part of railway engineering inspection. In order to address the challenge of timely detection of rail surface damage to prevent safety accidents and overcome the limitations of traditional detection methods, this paper proposes an automated damage segmentation algorithm that integrates multiple deep learning models, to provide a precise and efficient solution for the identification, analysis, and processing of rail surface damage.
Research conclusions: (1) A deep learning algorithm integrating target detection and semantic segmentation is proposed to achieve efficient recognition and accurate segmentation for rail surface damage features. (2) The practical results show that this algorithm has higher accuracy compared with several existing deep learning models, and the accuracy and average cross-merge ratio reach 99.56% and 83.89%, respectively. (3) This algorithm can accurately differentiate the blurred boundaries between rail damage and background, reduce data redundancy, and improve segmentation efficiency. It achieves high segmentation accuracy for multi-scale damage targets and the research conclusions can provide theoretical guidance for the automation of railway engineering inspection.
关键词
钢轨表面伤损 /
深度学习 /
图像分割 /
铁路工务
{{custom_keyword}} /
Key words
rail surface damage /
deep learning /
image segmentation /
railway engineering
{{custom_keyword}} /
中图分类号：
U291.69
{{custom_clc.code}}
({{custom_clc.text}})
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 田贵云,高斌,高运来,等. 铁路钢轨缺陷伤损巡检与监测技术综述[J]. 仪器仪表学报,2016(8):1763-1780.Tian Guiyun,Gao Bin,Gao Yunlai,etc. Review of Railway Rail Defect Non-destructive Testing and Monitoring[J]. Chinese Journal of Scientific Instrument,2016(8):1763-1780.
[2] Wang Weidong, Hu Wenbo, Wang Wenjuan, etc. Automated Crack Severity Level Detection and Classification for Ballastless Track Slab Using Deep Convolutional Neural Network[J]. Automation in Construction, 2021,124:103484.
[3] 刘蕴辉,刘铁,王权良,等. 基于图像处理的铁轨表面缺陷检测算法[J].计算机工程,2007(11):236-238.
Liu Yunhui, Liu Tie, Wang Quanliang, etc. Rail Surface Defects Detection Algorithm Based on Image Processing[J].Computer Engineering, 2007(11):236-238.
[4] 闵永智,岳彪,马宏锋,等. 基于图像灰度梯度特征的钢轨表面缺陷检测[J].仪器仪表学报, 2018(4):220-229.
Min Yongzhi,Yue Biao,Ma Hongfeng,etc. Rail Surface Defects Detection Based on Gray Scale Gradient Characteristics of Image[J].Chinese Journal of Scientific Instrument,2018(4):220-229.
[5] 贺振东,王耀南,刘洁,等. 基于背景差分的高铁钢轨表面缺陷图像分割[J].仪器仪表学报,2016(3):640-649.
He Zhendong,Wang Yaonan,Liu Jie,etc. Background Differencing-based High-speed Rail Surface Defect Image Segmentation[J].Chinese Journal of Scientific Instrument,2016(3):640-649.
[6] 袁小翠,吴禄慎,陈华伟.基于Otsu方法的钢轨图像分割[J].光学精密工程,2016(7):1772-1781.
Yuan Xiaocui,Wu Lushen,Chen Huawei. Rail Image Segmentation Based on Otsu Threshold Method[J].Optics and Precision Engineering,2016(7):1772-1781.
[7] 陈金胜. 基于图像识别的轨道定位及嵌入式实现[D]. 上海:上海工程技术大学,2017.
Chen Jinsheng. Track Location and Embedded Implementation Based on Image[D]. Shanghai:Shanghai University of Engineering Science,2017.
[8] Gan J, Li Q, Wang J, etc. A Hierarchical Extractor-based Visual Rail Surface Inspection System[J]. IEEE Sensors Journal, 2017(23):7935-7944.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
国家自然科学基金项目:基于数字孪生模型的轨道交通扣件系统伤损状态全生命周期演化机理研究(52178442);高速铁路基础研究联合基金项目:基于机器视觉的高速铁路基础设施服役状态智能监测理论及方法研究(U1734208)
{{custom_fund}}
PDF(1351 KB)
