基于机器视觉的多线路钢轨扣件缺损检测方法  

Rail Fastener Defect Detection Method for Multi Railways Based on Machine Vision

作　　者：刘俊博 黄雅平[1] 王胜春 赵鑫欣 邹琪[1] 张兴园 LIU Junbo;HUANG Yaping;WANG Shengchun;ZHAO Xinxin;ZOU Qi;ZHANG Xingyuan(Beijing Key Laboratory of Traffic Data Analysis and Mining,Beijing Jiaotong University,Beijing 100044,China;Infrastructure Inspection Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]北京交通大学交通数据分析与挖掘北京市重点实验室,北京100044 [2]中国铁道科学研究院集团有限公司基础设施检测研究所,北京100081

出　　处：《中国铁道科学》2019年第4期27-35,共9页China Railway Science

基　　金：国家自然科学基金资助项目(51827813);中国铁道科学研究院科技开发基金资助项目(2017YJ129);中央高校基本科研业务费专项资金资助项目(2019JBZ104,2016JBZ005)

摘　　要：提出基于在线学习策略的扣件区域定位算法,即根据轨道图像的先验知识和模板匹配方法定位扣件区域,利用在线学习策略动态地更新模板库,使算法能够在多线路的轨道图像中定位扣件区域,并自动标注扣件样本;提出基于深度卷积神经网络的扣件缺损识别算法,即根据扣件区域图像的尺寸较小、图像内容相对单一的特点设计识别算法的网络结构,采用样本随机排序策略的数据增强方法,以减小样本数量失衡对识别性能的影响。基于多线路钢轨扣件试验数据集对检测方法进行试验验证,结果表明:该方法可在不同线路的轨道图像中精确定位扣件区域并识别扣件缺损,扣件区域定位平均检测率达到99.36%,扣件缺损识别平均精确率达到96.82%,具有较高的可靠性和较强的多线路适应能力。Firstly,a fastener region location algorithm based on online learning strategy was proposed,which could locate fastener regions according to the prior knowledge of track image and template matching method.Online learning strategy was used to update the template library dynamically,so that the algorithm not only could locate fastener regions in the track images of multi railways,but also could automatically collect and annotate fastener samples.Secondly,a fastener defect recognition algorithm based on deep convolutional neural network was proposed.Namely,the network structure of recognition algorithm was designed according to the smaller size and the relatively single content of the fastener region image.The data augmentation method based on the sample random sorting strategy was adopted to reduce the impact of the imbalance of sample size on recognition performances.Test verification of the proposed method was conducted based on the rail fastener datasets of multi railways.Results show that the proposed method can locate fastener regions and identify fastener defects in the track images of different railways accurately,and it has high reliability and strong adaptability to multi railways.Specifically,fastener localization achieves an average detection rate of 99.36%and fastener defect recognition achieves an average precision of 96.82%.

关 键 词：钢轨扣件 缺损检测 多线路 图像识别 深度卷积神经网络 机器视觉 

分 类 号：U213.53[交通运输工程—道路与铁道工程]