Research On Track Surface State Recognition Method Based On Image Processing

The effective play of traction/braking performance of rail transit vehicle depends on the adhesion utilization of wheelset and track in contact with each other,and the change of track surface state is an important factor to change the adhesion between wheel and rail.When the rail transit vehicle changes from dry track surface to wet track surface,the adhesion coefficient between wheel and rail will decrease by half or more.The adhesion coefficient between wheel and rail decreased to the lowest when the dry track surface was changed to greasy track surface.At present,the state of track surface can only be judged by manual experience,and the identification efficiency is low.Therefore,this thesis takes the identification of track surface state as the research objective,combines image recognition and machine learning methods,and designs a track surface state recognition method based on image processing.The main research contents are as follows:(1)The method of track surface identification based on SVM fusion features is proposed.The visual information of track surface color and texture in different conditions was studied and analyzed,and the gray mean and variance of track surface image were calculated to describe the color features of track surface image,and the gray co-occurrence matrix was used to extract the texture features of track surface image.Then,the two characteristics are fused as the criterion of track surface state recognition,and the track surface state is recognized by SVM.(2)A deep learning based track surface state recognition method is proposed.Res Net-50 network was used to construct the track surface state identification model,and the structure and parameters of the network were optimized by transfer learning.The optimized RESNET-50 network was used to train the track surface state image data,and finally the model that could be used for classification was obtained to realize the track surface state recognition.And compared with the fusion feature SVM method.Track surface state change is the key factor that causes the change of the adhesion between wheel and rail.Based on image processing technology,this thesis uses the eature-integrated SVM method and deep learning method to identify the track surface state.Finally,the proposed method is verified by simulation experiment.