基于卷积神经网络的藏文手写数字识别  

Tibetan handwritten numeral recognition based on convolutional neural network

作　　者：夏吾吉 色差甲 扎西吉 贡保才让[1] 华却才让 XIA Wuji;SE Chajia;ZHAXI Ji1;GONGBAO Cairang;HUAQUE Cairang(Key Laboratory of Tibetan Information Processing of Ministry of Education,Qinghai Normal University,Xining 810008,China;Normal College for Nationalities,Qinghai Normal University,Xining 810008,China)

机构地区：[1]青海师范大学藏文信息处理教育部重点实验室,青海西宁810008 [2]青海师范大学民族师范学院,青海西宁810008

出　　处：《现代电子技术》2019年第5期79-82,86,共5页Modern Electronics Technique

基　　金：国家自然科学基金项目(61363055);青海师范大学科研基金项目(17ZR11);青海省科技计划项目(2017-GX-146)资助~~

摘　　要：藏文字处理和藏文数字自动识别等对藏文信息处理技术的潜在需求越来越高,目前已经成为藏区重要的研究课题之一。文中首先采集并构建藏文手写数字数据共15 000个样本,其中13 000个样本为训练数据,2 000个样本为测试数据,并对其进行预处理,后采用卷积神经网络(CNN)模型对藏文手写数字样本进行训练。经实验验证在测试集上的识别正确率达97.85%。The Tibetan information processing technology is potentially needed for Tibetan processing and Tibetan numeral automatic recognition increasingly,and has become one of the important research topics in Tibet.Tibetan handwritten numeral data is collected and constructed,which is 15 000 samples in total,in which 13 000 samples are taken as the training data,and 2 000 samples are taken as the test data and preprocessed.The convolutional neural network(CNN)model is used to train the Tibetan handwritten numeral samples.The experimental results show that the recognition correct rate of test set can reach up to 97.85%.

关 键 词：藏文手写数字 数字识别 CNN 数据预处理 样本训练 自动识别 

分 类 号：TN711.34[电子电信—电路与系统]