Research On The Application Of Freight Train Number Recognition Based On Image Processing In Marshalling Station

As a key part of the whole railway freight transportation system,marshalling stations are an important step to realize the information management of freight trains.However,the existing data show that the marshalling stations only identifie and manage trains at the entrance and exit,and the manual way is still used to record the number of freight trains in the marshalling stations,which has high work intensity and easy to produce error,and the distribution information of trains in each track in the stations is still blank.In view of the above research background,this thesis proposes a train number recognition method based on image processing,which installs an image acquisition device beside the track to record the train route.At the same time,the data sharing of various departments can be realized through the LAN.The staff can accurately grasp the distribution status of trains in the stations,timely modify the scheduling operation,and improve the operation efficiency of freight trains in the marshalling stations.The identification method of freight train number is studied.(1)Based on the analysis of train images in motion,this thesis analys various factors and challenges that need to be considered from the existing research results at home and abroad.According to the above analysis results,the recognition of railway train number is divided into three modules: train number regional positioning,character segmentation and single character recognition.(2)In the train number area positioning module,according to the railway freight train number marking standard,using the characteristics that the contents of the large and small train numbers are completely consistent and the target is small,the Faster R-CNN method is used to detect the large and small train numbers at the same time,and their contents complement each other.ZF,VGG16 and Res Net are analyzed and discussed as the basic feature extraction network to locate the big and small train number respectively,and the optimal network ZF is selected as the Faster R-CNN feature extraction network.Aiming at the problem of low positioning accuracy in traditional network,the traditional algorithm is improved in connection mode and suggestion box.In order to verify the effectiveness of the improved method,experiments are set up,and compared with the current positioning method.(3)In the train number segmentation module,the preprocessing method is used to process the image into a binary image,and then the ARLSA is used to segment the double line train number because of the foreground computing characteristics of the background pixels.In view of the problems of TILT and deformation of the train number characters,the undirected circle is used to segment the characters.The segmentation results show that character fracture will cause character segmentation error.Therefore,ARLSA is used to eliminate the fracture and optimize the segmentation results.(4)In the recognition module,GSO optimized BP neural network is used for recognition,it is called GSO-BP algorithm.Experimental results show that GSO-BP algorithm has higher recognition accuracy than traditional BP algorithm.