Automatic Processing And Calculation Of Feature Tags Of Track Images

As a necessary connecting part for fixing sleepers and rails,railway fasteners play an extremely important role in ensuring the safe operation of trains.Due to the increasing frequency of railway train operations,railway fasteners are prone to loosening,loss and other failures,which will cause incalculable serious consequences,so the status detection of fasteners is essential.However,the traditional fastener detection is manual,which is inefficient and average in accuracy.Therefore,according to the characteristics of the track image,automatic labeling and calculation of the label,maintenance of fasteners that are found to be in abnormal state in time,to ensure the high-speed and stable operation of the train,and thus to promote the country’s economic development has certain research significance.In order to better understand the characteristics of the track image,this paper first introduces the basic composition of the track structure and its spatial constraints,and classifies the characteristics of the image.In order to better extract the feature points,the traditional Harris-SIFT algorithm is limited by the coefficient k and the threshold T,which leads to the problem that the feature point extraction is not accurate enough and the calculation speed is slow.The Harris detection algorithm is improved while maintaining SIFT Scale-invariant feature transformation algorithm,an improved Harris-SIFT algorithm is proposed,and experiments show that the improved algorithm has strong real-time performance and high matching rate,and can also effectively reflect the shape characteristics of the track image.In order to effectively identify and judge the working status of fasteners,it is usually necessary to determine whether they are in a connected or non-fastened state.This paper adopts the method of pixel distance calculation and pattern matching to implement the state recognition of fasteners.Aiming at the shortcoming of the slow matching speed of the NCC algorithm in the feature area positioning,an improved NCC algorithm-FNCC is proposed.The basic idea is to reduce the amount of calculation in each matching process,so as to achieve the goal of ensuring accuracy and improving matching speed.And adopt the characteristic distance of the insulating cap and the characteristic distance of the elastic strip to judge the state of the fastener,set the fastening state of the fastener to the label state A,and the non-fastened(including damaged)state to B,through Simulation experiments show that the recognition rate of the two methods is very high,at the same time the combination of the two methods can help to make up for the missing,improve the reliability of state recognition,and reduce the detection rate to a large extent.Aiming at the problem of automatic identification and processing of fastener status labels,this paper introduces the composition and characteristics of convolutional neural networks,and studies the improved YOLOv3 fastener detection algorithm.Firstly,a deep feature extraction network is constructed using densely connected modules and transition modules,and five feature scale convolutional layer pyramids are designed to solve the problem that the scale of the feature map is large and the detection results are rough.Then the K-means algorithm is used to cluster the dimensions of the target frame again to make it better adapt to the characteristics of the fastener.Finally,the fastener data set is established and marked,and multiple sets of comparative experiments are set to test the algorithm.The results show that the improved YOLOv3 fastener detection algorithm has good robustness for fastener target detection and achieves better detection results in the presence of occlusion.