基于人工智能的脱机手写数字识别研究综述  

Summary of offline handwritten digit recognition research based on artificial intelligence

作　　者：张华美 张皎洁 ZHANG Huamei;ZHANG Jiaojie(College of Electronic and Optical Engieering&College of Microelectronics,Nanjing University of Post and Telecommunication,Nanjing 210023,China)

机构地区：[1]南京邮电大学电子与光学工程学院、微电子学院,江苏南京210023

出　　处：《南京邮电大学学报（自然科学版）》2021年第5期83-91,共9页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(61601242);江苏省博士后基金(1601150C)资助项目。

摘　　要：脱机手写数字识别技术是在光学字符识别技术的基础上,采用计算机等处理器自动对手写阿拉伯数字进行识别的一种技术。文中对国内外研究工作进行全面分析,首先介绍了基于人工智能技术的脱机手写数字识别历经的3个重要阶段:第一阶段是利用以支持向量机(SVM)为代表的传统分类器进行识别;第二阶段建立了卷积神经网络(CNN)为代表的神经网络模型;第三阶段设立了卷积神经网络和支持向量机相结合(CNN+SVM)为代表的混合分类模型。然后总结了人工智能技术在这3个阶段的优缺点,最后阐述了人工智能技术应用于脱机手写数字识别的问题和未来的发展方向。An affline handwritten digit recognition technology is proposed based on optical character recognition by using computers and other processors to automatically recognize handwritten Arabic numerals.According to a comprehensive survey of domestic and foreign researches,three important stages of the offline handwritten digit recognition technology based on the artificial intelligence are introduced as follows:1)the traditional classifier represented by support vector machine(SVM)is used for recognition;2)the neural network model represented by convolution neural network(CNN)is established;3)a hybrid classification model combined with the convolution neural network and the support vector machine(CNN+SVM)is introduced.Then,the advantages and disadvantages of the artificial intelligence technology in these three stages are summarized.Finally,the problems and the future development direction of the artificial intelligence technology applied to the offline handwritten digit recognition are expounded.

关 键 词：脱机手写数字识别 支持向量机(SVM) 卷积神经网络(CNN) 混合分类模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP183[自动化与计算机技术—控制科学与工程]