Research On Defect Recognition Of Rail Surface Based On Image And CNN Model

As part of the railway line that is in direct contact with the train,the rail guides and carries the train’s operation and also bears the great impact of the train.Therefore,the health of the rails directly affects the safety of train operations,especially the health of rail surfaces.Because the surface of the rail is directly in contact with the wheels of the train,it is a direct part of the impact force of the train.Therefore,timely detection of rail surface defects and repair and replacement is of great significance for ensuring safe running of trains,prolonging service life of rail and reducing economic losses.The existing artificial detection methods and ultrasonic testing methods can not meet the growing demand of China’s railway.Therefore,this paper combines image processing technology and Convolutional Neural Network(CNN,Convolutional Neural Network)model to detect the defects of the rail surface.The main contents of this thesis are as follows:First,the image of the rail surface was filtered.Combined with the advantages of Gaussian filtering and median filtering,and the characteristics of rail surface,we first reverse the image of the collected image,then use Gaussian-median filter to filter.The simulation experiment shows that the filtering method not only has good filtering effect,but also can protect and improve the details of image defect well.Second,extract the rail surface area.When collecting rail images,it was found that the rail widths on the same line are the same,and there are significant differences in gray levels between the rail surface area and the non-rail surface area.Therefore,Radon transform is used to project the rail surface image in the vertical direction,and the rail surface area is extracted through the projection curve and the track width.The simulation experiments show that the method can extract the rail surface area more completely.Finally,the CNN is used to classify the extracted rail surface image.Due to the limitation of conditions,this article mainly classifies the rail surface defects and the complete rail surface.First,the pre-processed pictures are input into the CNN model for training,and the best identification and classification model network is obtained through training.Then experiment with the trained network.The experimental results show that the CNN model has a good recognition rate for the rail surface dropping,further validates that the recognition and classification of rail surface defects is feasible based on image processing technology and CNN model.