基于图像识别的高铁接触网紧固件开口销故障分类方法

Image Identification-based Split Pin Fault Classification Method for OCS Fasteners of High Speed Railway

针对高铁接触网紧固件开口销缺失、松脱和安装不规范等现象,基于图像识别理论提出一种开口销故障等级划分方案和基于机器学习的开口销分类方法。在分类阶段,首先采用SSD算法对接触网4C系统的回图进行开口销定位与识别,并采用Deeplabv3+进行语义分割,最后采用SURF特征检测器对语义分割图片提取关键点,再利用视觉词袋模型BOVW生成视觉码本,利用视觉码本对极端随机森林ERF进行训练并生成模型。极端随机森林通过网格搜索和交叉验证(Gridsearch+CV)实现参数调优,使用ERF模型对开口销图片分类。采用该方法对实际线路的图片进行实验检测,准确率达到了93.2%,且节省人力,能有效保障高铁的供电安全。

With regard to the missing,looseness and unconformable installation of split pins of OCS fasteners for high speed railway,a scheme for split pin fault classification and a method for split pin classification are proposed on the basis of image identification theory.In the course of classification,firstly,return graphs of OCS 4C system will be positioned and identified by use of SSD algorithm,secondly,they will be segmented semantically by use of Deeplabv3+,thirdly,the key points will be extracted by use of SURF feature detector for generating of visual codebook by use of BOVW,the visual codebook will be used to train ERF for generating of a model.The ERF will achieve parameter adjustment and optimization by use of Gridsearch+CV and classify the graphs of split pins by use of ERF.The method is adopted to test and inspect the graphs of the actual railway line,with its accuracy reaching to 93.2%,saves the human resources and ensures the power supply safety for high speed railway.