Research On Train License Plate Recognition Based On Deep Learning

With the development of rail transportation in China,the requirement of train license plate recognition is becomes urgent.As an important part of intelligent transport system,government has paid great attention to it.Taking into account the complex diversity of train license plates,we will take train license plate recognition as natural scene text recognition.At present,the detection and recognition of texts in natural scenes based on deep learning have become a frontier topic in artificial intelligence and pattern recognition field.Therefore,train license plate recognition based on text detection and recognition technology in natural scenes has important practical and theoretical significance.The Train license plates in China do not have a unified standard forms,which are in various forms.This paper proposes three contributions based on the characteristics of train license plates.1.Discussion and analysis of two train license plate recognition frames based on deep learning.The first frame includes train license plate location based on traditional methods,train license plate character segmentation and Character recognition based on support vector machine/convolutional neural network.The second frame includes train license plate location based on deep learning,and train license plate sequence recognition based on recurrent neural network.2.This paper proposes a novel Binarization method based on fusion for characters of train license plate segmentation.3.This paper applies natural scene text detection and identification technology to train license plate detection and recognition.The entire article research on train license plate recognition based on deep learning technology,and two train license location methods have been proposed.One is based on edge detection,and the other is based on deep learning.The character segmentation method with insufficient prior information based on combination method of projection and connected-domain analysis are developed.Finally,for the character recognition of train license plates,this paper analyzes three character recognition methods including character recognition based on support vector machine,character recognition based on convolutional network and character sequence recognition based on recurrent neural network.At present,the automatic license plate recognition algorithms in this paper have been applied in more than 30 railway section,such as Guiyang section of Cheng du railway bureau,Huaihua section of Guangzhou railway bureau and Hangzhou section of Nanjing railway bureau.In train licenses recognition system,recognition rate is 85%.The recognition rate is over 92% combining the results of RFID(Radio Frequency Identification).