基于卷积神经网络的轨道表面缺陷检测技术实现

Implementation of Track Surface Defect Inspection Based on Convolutional Neural Network

利用计算机视觉技术,通过采用基于线阵CCD相机的图像采集系统,能够实现轨道信息的完整采集,实现对铁路轨道缺陷的检测。然而,目前国内大部分缺陷识别技术还不成熟,对于缺陷数据的自动化识别仍很困难,很大程度仍然依靠高度熟练的技术工人或者采用进口检测系统。因此,为满足发展需求,针对有效的轨道表面缺陷识别检测方法进行研究,提出基于卷积神经网络的算法,实现对轨道表面缺陷的识别。以圆形、条形两种缺陷类型数据作为研究对象,验证对轨道表面缺陷的识别检测准确度。

Based on computer vision technology, with the image acquisition system based on linear CCD camera, the comprehensive track information can be collected to lay the foundation for defects detection and location, thereby achieving the inspection of railway track defects. While the most current domestic defect recognition technology has not been refined, and there are setbacks automatic identification of statistics which largely relies on skilled workers or imported detection systems. With this regard, conducts research on the effective method in track surface defects recognition and detection to meet the demand in railway development, and successfully proposes an algorithm based on convolution- al neural network to achieve the recognition of the track surface defects. What＇ s more, the accuracy of the identification of orbital surface defects is verified by researching circular and strip, the two types of defects statistics.