Research On Rail Surface Defect Detection Technology Based On Image Recognition

The safety and stability of railways play an important role in the development of our country’s economy.Traffic accidents caused by rail defects have caused extremely serious consequences for life safety and social property.Therefore,it is of great practical significance to study accurate and efficient rail surface defect detection technology to detect defects and repair them in time before accidents occur.The method of using image processing technology to detect rail surface defects has the advantages of high accuracy and non-contact,and has become an important method in today’s rail surface defect detection.Based on the comprehensive consideration of detection accuracy and classification speed,this paper solves the existing problems in the existing rail surface defect detection algorithm,and focuses on the research of the rail surface defect detection algorithm based on machine learning and deep learning.The main research contents are as follows:(1)Aiming at the slower problem of median filtering algorithm,design mean and median fusion filtering algorithm.Aiming at the problem that the collected image contains non-rail areas,which affects the recognition accuracy,a rail surface extraction algorithm based on Line Segment Detector(LSD)is proposed to filter out the complicated background part.The defective part of the railway line is relatively small.In order to avoid the detection of a large number of defect-free images and increase the detection speed,the cache mechanism is introduced to improve the difference hash algorithm,and the initial defect judgment algorithm is designed.(2)Design a rail surface defect detection algorithm based on machine learning.Aiming at the slow processing speed of the two-dimensional Otsu image threshold segmentation algorithm,the Particle Swarm Optimization(PSO)algorithm is improved,and the PSO-based image segmentation algorithm is designed to speed up the segmentation processing speed.Aiming at the problem that a single feature is insufficient to describe the image information of the rail surface defect,multiple features of the defect area are extracted and merged,including gray-scale features,shape features,and texture features.In view of the problem that the dimensionality of the data is too large after multi-feature fusion,which leads to the slow classification speed,the Principal Components Analysis(PCA)is used to reduce the dimensionality of the fused features,and a Support Vector Machine(SVM)is used to construct a multi-classifier to identify and classify defects.(3)Design a rail surface defect detection algorithm based on deep learning.Aiming at the problem that the feature extraction module in the Faster R-CNN(Region-based Convolutional Neural Networks)algorithm does not fully extract low-level semantic information,and thus cannot fully describe the defect details,a multi-scale feature fusion extraction method is adopted.Aiming at the problem of missed detection of small-size defects,improve the size and number of anchor windows in the Region Proposal Networks(RPN),and design a weighted ROI pooling method to increase the detection accuracy of small-size defects.Combining the types of defects in the task of this article,optimize the output structure of the Faster R-CNN algorithm,and determine the training method of the neural network.Finally,a dataset of rail surface defects is established,and two different detection algorithms based on machine learning and deep learning are experimentally verified.The results show that the detection accuracy of the two algorithms are 92.7% and 93.9%,respectively,but the detection based on deep learning The running time of the algorithm is significantly better than the detection algorithm based on machine learning.