摘要
为提高轨道扣件状态检测的准确率,基于K均值聚类算法改进掩膜区域卷积神经网络(Mask R-CNN)实例分割算法中的区域建议网络。进行基于改进Mask R-CNN的轨道扣件状态检测方法研究,并将该方法分别应用于普速铁路有砟轨道2个扣件数据集和高速铁路无砟轨道1个扣件数据集上进行轨道扣件状态检测。结果表明:该方法能对普速铁路有砟轨道和高速铁路无砟轨道图像中的扣件状态进行准确检测,扣件的定位准确率和分类准确率平均分别达到97.05%和98.36%,均优于YOLO V3,Faster R-CNN和Mask R-CNN算法;相较于前2种算法,本方法对普速铁路有砟轨道扣件状态检测的优势更为明显。
In order to improve the accuracy of track fastener state detection,the region proposal network in Mask Regional Convolutional Neural Network(Mask R-CNN)instance segmentation algorithm was improved based on K-means clustering algorithm.The method of track fastener state detection based on improved Mask R-CNN was studied,and the method was applied to track fastener state detection on two fastener data sets of the ballast track in conventional speed railway and one fastener data set of the ballastless track in high-speed railway.Results show that the method can accurately detect the fastener state in the images of the ballast track in conventional speed railway and the ballastless track in high-speed railway.The average positioning accuracy and classification accuracy of fasteners reach 97.05%and 98.36%,respectively,which are better than YOLO V3,Faster R-CNN and Mask R-CNN algorithm.Compared with the former two algorithms,this method has more obvious advantages in detecting the fastener state of ballast track in conventional speed railway.
作者
许贵阳
李金洋
白堂博
杨建伟
XU Guiyang;LI Jinyang;BAI Tangbo;YANG Jianwei(School of Mechanical-Electronic and Vehicle Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Beijing Key Laboratory of Performance Guarantee on Urban Rail Transit Vehicles,Beijing University of Civil Engineering and Architecture,Beijing 100044,China)
出处
《中国铁道科学》
EI
CAS
CSCD
北大核心
2022年第1期44-51,共8页
China Railway Science
基金
国家自然科学基金资助项目(51975038)
北京市自然科学基金资助项目(KZ202010016025)。
关键词
轨道
扣件
状态检测
掩膜区域卷积神经网络
K均值聚类算法
定位准确率
分类准确率
Track
Fastener
State detection
Mask regional convolutional neural network
K-means clustering algorithm
Positioning accuracy
Classification accuracy