基于多特征融合的地铁车辆制动组件异常检测

Anomaly detection of metro vehicle brake components based on multi feature fusion

地铁车辆驾驶环境多变,导致制动组件异常检测存在精度误差,为此提出基于多特征融合的地铁车辆制动组件异常检测方法。通过无人机与云台搭载相机,采集地铁车辆制动组件运行图像。通过Gabor特征提取方法提取组件图像空间方向与尺度上的多种纹理特征。采用信息熵实现地铁车辆制动组件图像多个提取特征的融合。基于CNN设计BD-YOLO地铁车辆制动组件异常检测模型,实施制动组件异常检测。测试结果表明,在实验地铁车辆静止时,该方法的组件异常检测精确率达到了100%。在车辆正常运行的情况下,其组件异常检测精确率较高。在正常运行中列车管不充风的情况下,其组件异常检测宏平均召回率整体高于95%。

Due to the changeable driving environment of metro vehicles,there are precision errors in brake component anomaly detection.Therefore,an anomaly detection method for metro vehicle brake components based on multi feature fusion is proposed.The UAV and PTZ are equipped with cameras to collect the running images of the brake components of metro vehicles.Gabor feature extraction method is used to extract multiple texture features in the spatial direction and scale of component image.Information entropy is used to realize the fusion of multiple extracted features of subway vehicle brake component image.Based on CNN,BD-YOLO metro vehicle brake component anomaly detection model is designed to detect brake component anomaly.The test results show that the accuracy of component anomaly detection reaches 100%when the experimental metro vehicle is stationary.Under the normal operation of the vehicle,the accuracy of its component anomaly detection is high.Under the condition that the train pipe is not filled with air during normal operation,the macro average recall rate of its component anomaly detection is higher than 95%as a whole.