A Railway Fastener Detection Method Based On Modified Faster R-CNN

Since the 21 st century,China’s railway construction has developed rapidly,and has entered the era of high-speed railway.With the continuous expansion of the scale of high-speed railway,in order to ensure the safe and smooth operation of high-speed railway,testing of high-speed railway track has become the premise and basis for the development of high-speed railway.High-speed railway fasteners play a role in fixing the railway,maintaining the gauge,stopping the railway from moving longitudinally and horizontally relative to the railway sleeper,when the high-speed railway fasteners on the track deviation,loss,easy to cause the high-speed railway running process body instability,and then lead to accidents,causing great losses to people’s lives and property safety.Therefore,for the sake of make the high-speed railway run safely,it is of great significance and value to conduct accurate and rapid inspection of high-speed railway fasteners.Traditional fastener inspection uses manual inspection,which is time-consuming,costly and involves safety risks.With the development of computer vision technology,the technology is widely used in the detection of high-speed railway fasteners,and this method is safer and faster than manual inspection.At the same time,with the development of image signal processing algorithm,the target detection algorithm based on deep learning is gradually applied to the detection of high-speed railway fasteners in the process of detection of high-speed railway fasteners.In order to further improve the detection accuracy and detection speed of the detection network,and to address the current problems of loss and deviation of high-speed railway fasteners,this thesis proposes an improved Faster R-CNN based detection method for high-speed rail fasteners,and the main research contents are as follows:(1)Construction of high-speed railway fasteners dataset.In this thesis,a series of high-speed railway fastener images were collected using the experimental site of the State Key Laboratory of Railway Transit Infrastructure Performance Monitoring and Assurance,and the high-speed railway fastener images were expanded by data enhancement techniques to construct the high-speed railway fastener dataset.(2)In order to improve the detection accuracy of high-speed railway fasteners for multiple types of different states,this thesis adopts a detection network model based on the improved Faster R-CNN for the detection of high-speed railway fasteners.First,deformable convolution is introduced in the feature extraction network Res Net-101,and an offset is added to the position of each sampling point in the convolution kernel to make it more focused on the fastener region in the feature extraction process and achieve the accurate extraction of fastener features;Secondly,to address the problems that the regression loss Smooth L1 in Faster R-CNN is insensitive to the condition that the predicted edge overlaps with the real edge and unfavorable to the predicted edge regression,Alpha-IOU loss is used to optimize the Faster R-CNN network model instead of Smooth L1 loss.The experimental results show that the improved high-speed railway fastener detection model can effectively detect two different states of multiple types of high-speed railway fasteners lost and deviate,and improve the detection accuracy of the detection network model.(3)In order to reduce the computation of the detection network and improve the detection speed of high-speed railway fasteners,two improved lightweight detection network models,namely Shuffle R-CNN and lightweight Faster R-CNN,are used in this thesis.The experimental results show that both methods effectively improve the detection speed while guaranteeing certain accuracy.