Multi-model Rail Surface Defect Detection Method Based On Convolutional Neural Networks

Because the railway is energy-saving and convenient.At present,it has become one of the most important means of transportation for people to travel.In addition,compared with other transportation industries,the safety of railway transportation is very high.At present,with the prosperity of China’s railway industry and the increase of operating mileage,speed and density,the requirements for rail detection are further improved.At the same time,with the progress of science and technology and the rise of computer network,new technologies such as big data,machine learning and so on have been developed continuously.Deep learning emerges as the times require,gradually breaking through the bottleneck of traditional methods.However,it is still a difficult step to combine deep learning with railway rail to achieve accurate detection of rail defects.Therefore,aiming at the railway rail,this paper studies the establishment of the detection system of the rail squat defect and the recognition of the deep learning method,proposes a method of multi model fusion,which realizes the rapid and accurate detection of the defect,and proves that the proposed method can effectively improve the safety of Railway transportation and passengers through the comparative analysis of experiments.First of all,the paper briefly introduces the research background and significance of rail defect detection,analyzes the current research difficulties of rail surface defect detection,introduces the traditional physical detection methods and machine vision methods currently used at home and abroad,and finally summarizes the content of this paper.Secondly,the paper analyzes the problems encountered in the current rail defect detection system at home and abroad,puts forward the rail surface defect detection system,and describes the composition and function of the system in detail.Then,the paper focuses on two kinds of rail surface defect detection algorithms:SSD algorithm and yolov3 algorithm.After introducing the principle of the original algorithm and the network,starting from the problem of rail surface defect detection,in order to realize the detection of micro,medium and large rail black spot defects,the two algorithms are improved,and compared with the original algorithm in terms of detection accuracy,running speed and other performance aspects,so as to prove the superiority of the proposed algorithm.Finally,taking the proposed rail surface defect detection system as the carrier and two improved rail black spot defect detection algorithms as the core,the multi model fusion rail surface defect detection method based on convolutional neural network is tested in five different scenarios,and compared with a variety of main flow algorithms to prove the robustness of the proposed method.In this paper,the SSD and yolov3 detection algorithm are improved.By using the method of multi model fusion,the two algorithms can detect the rail defects in parallel.Without losing the detection speed,the accuracy of rail defect identification is improved and the normal operation of the railway is better guaranteed to some extent.