Research On Online Mongolian Handwriting Recognition System Based On Deep Learning

The mainstream Mongolian input method is based on keyboard input,and has matured in product development.However,the Mongolian input method for handwriting recognition input method is still in its infancy.Handwriting recognition technology is divided into online recognition and offline recognition.This thesis develops a set of online Mongolian handwriting recognition system based on deep learning technology for smart phone terminal development.This thesis draws on the theoretical basis and technical route of online handwriting recognition model in mainstream languages such as Chinese and English,and builds an online Mongolian handwriting recognition model based on sequence-to-sequence(Seq2Seq)framework.Firstly,this paper develops a Mongolian handwritten data acquisition system based on Android.By collecting and collating the Mongolian handwritten corpus,257,800 samples(total vocabulary of 124,401)were finally used to participate in model training.Secondly,this thesis proposes a pre-processing of Mongolian handwritten recognition data include distance angle method and coordinate standardization method.Finally,the online Mongolian handwriting recognition model based on deep learning technology is free from the dependence of data feature extraction.The powerful modeling ability of the sequence-to-sequence(Seq2Seq)model improves the performance of handwriting recognition.The experimental results show that the correction rate of the Mongolian word recognition system based on Seq2 Seq model reaches 83.18%,and it is proved that the data preprocessing operation can further reduce the calculation amount and improve the recognition efficiency.Finally,based on the online Mongolian handwriting recognition model of Seq2 Seq model,this thesis builds a complete Mongolian handwriting recognition service,and realizes the development of handwritten Mongolian input method by calling this service through smart phone terminal.