Research On Chinese Character Recognition Technology Based On Deep Learning

Chinese character recognition is an important branch of pattern recognition research and application field.It has been widely applied in many fields,such as economy and trade,intelligent transportation,character printing,and so on.Therefore,more and more attention has been paid to the research and development of Chinese character recognition methods and technologies.Nowadays,it has become one of the most popular research topics in the world.With the rapid development of deep learning in the field of artificial intelligence,convolutional neural networks in deep learning have achieved excellent recognition performance in pattern recognition.Due to the limitation of feature extraction method in traditional Chinese character recognition,convolution neural network in deep learning plays the advantage of automatically acquiring the characteristics of learning samples.But deep learning is more complex when training samples,so it is more difficult.This paper studies the application of convolution neural network and traditional pattern recognition in the application of Chinese character recognition.The main research work of this paper is as follows:1)In this paper,the traditional Chinese character recognition method is studied.The support vector machine plus decision tree is used as classifier.We use directional feature,Gabor feature and elastic network feature together with three different data sets preprocessing methods.The experiment shows that the data set preprocessing method using the morphologic conversion and the feature of the elastic network can obtain better recognition accuracy.2)In order to solve the problem of that the minor differences among the Chinese characters of similar shape will be lost in the training process,the weight of small differences at the Chinese characters image were increased through the convolution layer of the selected AlexNet network and the attention layer in parallel.The results show that the added attention layer can reduce the impact of the deep layers for losing information,thereby improving performance.3)During the implementation of Chinese character recognition system,data enhancement is studied.After data set is extended by reasonable data enhancement method,experiments show that it can avoid overfitting of deep convolution neural network effectively.4)Combined with the results of data enhancement and data set preprocessing,experiments on convolution neural networks with different structures are carried out.Among them,the accuracy of Chinese character recognition using the method of data enhancement plus morphological transformation plus AlexNet integrated attention layer is 99.87%,which indicates that the application of this method in Chinese character recognition is effective.5)Combined with feature extraction and convolutional neural network in traditional image recognition,using elastic network feature map as input of convolution neural network,can reduce volume accumulation and improve recognition effect,and optimize network performance.