Research On Image-Based Detection And Identification Technologies Of Rail Surface Defects

The safety of train running track is very important to the safe running,so the defect detection of running track is of great significance to the safe and efficient running of train.The traditional detection methods of track damage,such as visual method,magnetic powder method and ultrasonic detection lamp method,have low detection efficiency,are susceptible to subjective factors with high cost and low accuracy.In order to solve the above problems,a method of rail defect detection based on machine vision came into being,which combines the rail surface image collected by rail inspection equipment with image processing technology to detect defects.This method has the advantages of high efficiency,convenience,low cost,and a broad application prospect.In this paper,based on the research background of rail surface image,combining with the principle of machine vision and deep learning.An image-based method of rail surface defect recognition is proposed.Firstly,the research background and significance are expounded,the important role of rail surface defect detection is discussed,the research status is reviewed and analyzed,and the existing research is summarized.According to the detection requirements of the rail inspection image,the hardware of the image acquisition device of the detection system is selected,and the theoretical research and technical analysis of the parameter settings are carried out to obtain a good rail surface image.Then,the collected track image is preprocessed,and the histogram equalization algorithm is used to enhance the track image.Then,several commonly used image denoising algorithms are compared,and the adaptive filtering algorithm is used to denoise the track image.After the preprocessing,the improved Canny edge detection operator is used to roughly extract the defect image after preprocessing.In this paper,we introduce the morphological principle to reconstruct and filter the background texture,and then use the multi constraint Hough transform to detect the straight track,and get a better straight track model extraction results.By comparing the edge detection algorithm and threshold segmentation algorithm,the Ostu threshold segmentation algorithm is improved to complete the defect segmentation module.Finally,using the supervision way to adjust the weights,the neural network weights according to the features of loss to learn all kinds of defects,the features of the effective information rights,invalid or effect,the feature information of the small weight is small,in the way to train network model,constructing DenseNet and SE-DenseNet network respectively,after the split defect images,defect classification,get a better recognition effect.The experiment shows that the rail surface defect recognition technology based on image proposed in this paper can quickly and accurately locate the rail area and the existing defects;accurately classify and identify the defects,which can meet the requirements of the accuracy and speed of rail inspection image detection,has high theoretical value and practical application prospects.