基于深度卷积神经网络的脱机手写汉字识别系统的设计与实现  

Offline Handwritten Chinese Character Recognition Based on Convolutional Neural Network

作　　者：林恒青[1] LIN Hengqing(Department of Mechanical Engineering,Fujian Chuanzheng Communications College,Fuzhou Fujian 350007)

机构地区：[1]福建船政交通职业学院机械工程系,福建福州350007

出　　处：《湖北理工学院学报》2019年第2期31-34,共4页Journal of Hubei Polytechnic University

基　　金：2017年福建省教育厅科技项目(项目编号:JAT170942)

摘　　要：针对传统手写汉字识别的特征提取方法的局限性,采用深度学习的方法对手写汉字进行图像识别,借鉴GoogLeNet网络搭建了一个适合手写汉字识别的卷积神经网络。利用BN层训练神经网络,最终训练得到的神经网络在验证集上的top1正确率达到96%以上,top3正确率达到98%以上,并具有很好的泛化能力。这表明深度学习为脱机手写汉字识别系统的设计提供了新的思路,具备自动提取特征,且具有一定的实际意义。Aiming at the limitation of traditional feature extraction methods for handwritten Chinese character recognition,image recognition for handwritten Chinese is done with the deep learning method and a convolution neural network suitable for handwritten Chinese character recognition is built according to Goog Le Net.Batch Normalization Layer is used to train the neural network.The correctness rate of Top1 on the verification set is more than 96% and that of top3 is more than 98% .The neural network has good generalization ability.The results show that deep learning provides a new idea for the design of offline handwritten Chinese character recognition system,which has the feature of automatic feature and the practical application.

关 键 词：手写汉字识别 深度学习 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]