Research On Recognition Of Train Wheel Tread Damage Based On Image Processing

The damage of train wheel tread will directly affect the safety of train operation and threaten the safety of passengers' life and property.Therefore,more and more attention has been paid to the detection and identification of wheel tread.At present,the detection of train wheel tread damage in China is mainly based on manual inspection.This method relies heavily on the technical proficiency and on-site experience of overhaul workers.The method is less efficie nt and has a higher missed rate.After deeply researching the train wheel tread damage detection system at home and abroad,this thesis focuses on the application of digital image processing method in wheel tread damage detection,and also provides technical support for the automatic detection of wheel tread damage.This thesis mainly studies from the following aspects:(1)This dissertation analyzes and researches the existing tread image acquisition systems at home and abroad,and determines the final image acquisition system based on this.The system includes cameras,light sources,speed sensors and trigger sensors.The device completes the collection of tread images in the train overhaul workshop.(2)The collected tread image is filtered and enhanced,and then several common edge detection algorithms and the improved Canny algorithm proposed in this thesis are used to perform edge detection on the image.The improved Canny algorithm uses an adaptive weighte d median filter algorithm to smooth the image and the Otsu method determines the best high and low thresholds.(3)Pixels are selected on the rim edge line after edge detection.Then the least squares method is used to fit the left and right edge lines,and the tread area within the left and right edge lines is preserved and the area is segmented on the original image.After the binariza t io n and morphological operations are performed,the suspicious damage area is found.(4)The feature quantity analysis is performed on the suspicious damaged area and the texture feature is obtained by calculating the gray level co-occurrence matrix.Several Haralick features with high classification accuracy are selected,namely energy,contrast,entropy,correlation and inverse difference matrix,and then BP neural network is used for classifica t io n and recognition.The simulation experiment results show that the improved Canny algorithm can effective l y detect the edge of wheel tread and keep the true edge of the train wheel tread.The image processing method can efficiently identify the damage area.The BP network is used to identify and classify features,and the effect is better.