Research On Classification And Recognition Of Train Wheelset Tread Defects Based On Image Processing

With the rapid development of the rail transit industry and the development of heavy loads,the safety of train operation has attracted widespread attention.The train wheel tread is the part of the running part that directly contacts the surface of the steel rail,and its defects will bring safety hazards to the train operation.Therefore,the accurate determination of the tread state is the basic guarantee for the stable development of the railway industry.The thesis mainly uses the method of combining image processing and convolutional neural network to realize the classification and recognition of wheelset tread defects.The main research contents are as follows:(1)Image preprocessing of wheelset tread.The weighted average method,bilateral filtering,and power law transformation are used to preprocess the original wheel tread image in sequence.The image processed by the above algorithm obtains a higher Peak Signal-to-Noise Ratio(PSNR)and a lower Mean Square Error(MSE),and the processed image quality can meet the processing requirements.(2)Tread edge detection and fine segmentation.An improved Canny edge detection algorithm is proposed.From the simulation results,the improved algorithm is significantly better than the traditional algorithm in terms of edge continuity and noise filtering.The edge line is fitted by the least square method,and the tread area is accurately segmented according to the fitted line.(3)Tread defect area location and defect texture feature extraction.A binarization segmentation method based on block local threshold is proposed.In order to obtain better connected regions,mathematical morphological operations are performed on the binarized image,and then all connected regions are labeled with the smallest bounding rectangle.Through the mathematical feature screening of the defect,the suspicious tread defect area is finally obtained;the gray-gradient co-occurrence matrix is used to describe the tread texture feature,and 15 texture feature quantities are extracted.(4)Classification and identification of wheel tread defects.According to the extracted tread texture feature quantity,the small gradient advantage,energy and inertia of the texture feature quantity are selected,and the K-means clustering method is used to cluster four types of wheel tread.The simulation results show that the classification effect of each type of tread is more than 95%.An improved Alex net convolution neural network method is proposed to identify the defects of wheelset tread,and the recognition rate of tread defects is 94.1%.