摘要
传统人工巡检轨道交通缺陷存在效率低、误差大等缺点,运用综合轨道检查车(CTIV)为检测平台,建立了一种车载式轨道交通图像识别智能巡检系统。利用CTIV采集连续轨道边界框图像建立数据集,在C++中使用应用程序开发框架(QT)设计了可视化数据标注的定制软件工具。通过全卷积网络(FCN)建立了多任务学习扩展网络架构,结合多个检测器可以提高轨枕和轨道弹力紧固件的缺陷检测性能。实验结果表明,提出方法在检测轨枕和紧固件缺陷时具有较高的精度。
The traditional manual inspection of rail transit defects has the disadvantages of low efficiency and large error.A vehicle mounted rail transit image recognition intelligent inspection system is established by using the comprehensive rail transit inspection vehicle(CTIV)as the detection platform.CTIV is used to collect the image of continuous track boundary frame to establish data set,and QT is used in C++to design a customized software tool for visual data annotation.The multi task learning extended network architecture is established by full convolution network(FCN),and the defect detection performance of sleeper and track elastic fastener can be improved by combining multiple detectors.The experimental results show that this method has high accuracy in the detection of sleeper and fastener defects.
作者
马茜
MA Qian(Xixian New District Rail Transit Development Co.,Ltd.,Xi'an,Shaanxi 710000,China)
出处
《计算技术与自动化》
2022年第1期117-122,共6页
Computing Technology and Automation
关键词
轨道交通
图像识别
全卷积网络
轨枕
紧固件
rail transit
image recognition
full convolution network
sleeper
fastener
