Research On Rail Surface Defect Recongnition Based On Convolutional Neural Network

With the rapid development of China’s railways,rail surface defects are closely related to the safety of railway transportation.Therefore,it is very important for the detection of rail surface defects.At present,the traditional detection methods are difficult to meet the precise and efficient detection requirements.Because convolutional neural networks can automatically learn the characteristics of sample features,they have a great advantage in image processing compared to other deep learning mo dels.RCNN,Fast RCNN,and Faster RCNN network structure models based on convolutional neural network model have made great progress in target detection than traditional algorithms.First,this paper studies the deep learning theory,especially the convolu tional neural network in the field of target detection.Then it compares the detection performance of RCNN,Fast RCNN and Faster RCNN network structure models.Based on deep learning and based on the network structure of the convolutional neural network,t he paper optimizes the Faster RCNN network model from the aspects of feature extraction network,anchor boxes,output adjustments,and experimental parameter settings.And then use the optimized and improved network model to detect the rail surface defects,and ultimately can accurately and efficiently detect the rail surface defects.This paper compares the extraction of candidate regions by Region Proposal Network and Selective Search algorithm,and verifies that the quality of candidate regions extracted by Region Proposal network is superior to Selective Search algorithm.At the same time,in the network structure model of Faster RCNN,it is difficult to effectively detect the smaller size of the rail image.Through the surface of the experiment results,the improved anchor window setting can accurately detect the surface defects of the rail with smaller size.The experimental results shows that the detection performance of the improved Faster RCNN network structure model is better than the detection method based on image processing and the detection method based on Faster RCNN network model both in the average detection accuracy and the average detection time.In the whole rail surface defect detection,the method used in this paper achieves better result s in terms of detection accuracy and real-time performance.