Railway Train Number Recognition System Based On Deep Learning

With the rapid development of China's railway industry,the daily maintenance of railway trains are becoming more and more important.In order to realize the information management of inbound maintenance trains,accurate recognition of train number is needed as a guarantee to improve the management efficiency of railway trains.Because manual recording and checking of train number information require a lot of manpower and material resources,the electronic tag technology based on RFID is not easy to maintain and costly,and the accuracy of traditional image recognition methods is not high,so this thesis uses the deep learning technology which is outstanding in image recognition to study the train number recognition.In this thesis,the train number recognition is mainly based on the deep learning object detection algorithm,in order to avoid the problems caused by the more intermediate processes in the traditional recognition method and to simplify the working process and improve the accuracy of the recognition system.Firstly,the sample data for training and testing are obtained by sorting out a large number of original train number image sets collected on site,and the original images are labeled manually.Secondly,the train number region is detected and located based on the deep learning object detection algorithm,which overcomes the limitations of traditional image location method.And the train number region is used for train number character detection and recognition.Then,train number characters are also detected based on deep learning object detection algorithm to achieve end-to-end recognition of train number characters without the task of character segmentation.Finally,the train number location and character detection are combined to realize the train number recognition system.Through testing,the high recognition accuracy of the system is verified,and the practical application is carried out,which solves the problems of high equipment cost and insufficient reliability of traditional recognition methods.