摘要
繁忙的铁路运输会导致钢轨表面病害的出现,及时对钢轨表面进行检测能有效确保铁路的安全运行。首先介绍了钢轨表面病害的产生原因以及病害类型;其次分析了自动检测技术中机器视觉检测法的实现系统以及该方法的优势和面临的困境;再次从轨道背景、病害分割提取以及病害分类评估3个方面归纳了现如今主流的钢轨表面病害识别处理方案,即传统图像方法和机器学习方法。传统方法从纹理、图像语义2个方面进行特征提取和病害识别的算法设计,实现了对病害的识别、提取和预警,通过优化算法来克服环境因素带来的干扰,提高鲁棒性。近年来研究趋向于将传统方法与机器学习方法相结合,通过深度学习模型自动学习特征,弥补传统方法的不足,发挥机器学习方法在自适应性和泛化能力方面的优势,以实现更高的性能和应用广泛性。2种方法在特征设计方式、数据驱动与人工设计差异以及适应性和通用性3个方面各有优缺点。最后,提出视觉检测技术未来的发展方向应主要集中于提高算法的兼容性和多数据融合能力方面。
Busy railway transportation will lead to rail surface diseases,timely detection of rail surface can effectively ensure the safety of railway operation.Firstly,the causes and types of rail surface diseases were introduced.Secondly,the realization system of machine vision detection method in automatic detection technology was analyzed,as well as the advantages and difficulties faced by this method.Thirdly,the current mainstream recognition and processing schemes,namely traditional image method and machine learning method,were summarized from three aspects of track background,disease segmentation and extraction,and disease classification and evaluation.Traditional methods design feature extraction and disease recognition algorithms from texture and image semantics to realize disease recognition,extraction and early warning,and optimize algorithms to overcome the interference brought by environmental factors and improve robustness.In recent years,there has been a trend to combine traditional methods with machine learning methods,to make up for the shortcomings of traditional methods through automatic learning features of deep learning models,and to give play to the advantages of machine learning methods in adaptability and generalization ability to achieve higher performance and application universality.The two methods have their own advantages and disadvantages in feature design,differences between data-driven and manual design,and adaptability and universality.Finally,it is proposed that the future development direction of visual inspection technology should focus on improving the compatibility of algorithms and multi-data fusion ability.
作者
何成刚
张坤雄
俞茹昕
王欣纪
徐逸勋
刘吉华
HE Chenggang;ZHANG Kunxiong;YU Ruxin;WANG Xinji;XU Yixun;LIU Jihua(School of Rail Transportation,Wuyi University,Jiangmen Guangdong 529020,China;Key Laboratory of Intelligent Operation,Maintenance and Emergency Management for Rail Transit of Jiangmen(Wuyi University),Jiangmen Guangdong 529020,China)
出处
《高速铁路新材料》
2024年第1期7-13,共7页
Advanced Materials of High Speed Railway
基金
广东省教育厅特色创新项目(2023KTSCX151)
广东省级大学生创新创业训练计划项目(S202211349073)
五邑大学创新创业基金项目(2023111500000421)。
关键词
钢轨表面病害检测
机器视觉
深度学习
rail surface disease detection
machine vision
deep learning
