基于深度学习的数字识别模块在安卓系统的实现  

Realization of a Digital Recognition Program Based on Deep Learning in the Android System

作　　者：梁淑芬[1] 胡帅花 秦传波[1] 邓志强[1] LIANG Shu-fen HU Shuai-hua QIN Chuan-bo DENG Zhi-qiang(School of Information Engineering, Wuyi University, Jiangmen 529020, Chin)

机构地区：[1]五邑大学信息工程学院,广东江门529020

出　　处：《五邑大学学报（自然科学版）》2017年第1期40-45,共6页Journal of Wuyi University(Natural Science Edition)

基　　金：广东省特色创新类(教育科研类)资助项目(2015GXJK151);广东省教育厅特色创新类资助项目(2015KTSCX148);广东省江门市资助项目(江科[2014]145号)

摘　　要：针对手写数字、印刷体数字和身份证号码识别问题,研究了深度学习卷积神经网络算法,并在安卓系统上开发了数字识别应用模块,实现了数字图像的采集、裁剪和定位等预处理,并运用卷积神经网络的Le Net网络模型进行数字识别.实验结果表明,本文实现的数字识别程序使用简便,对手写数字、印刷体数字和身份证号码识别率分别为94.77%、98.32%和99.78%,识别率较高,具有很好的推广前景.In order to solve the problem of identifying handwriting, print numbers and ID numbers, the deep learning convolutional neural network（CNN） algorithm is studied, and an App of digits recognition system is developed on the Android smart phone platform to achieve preprocessing such preprocessing as digital image acquisition, cutting and positioning.Also the Le Net network model of the convolution neural network is used to conduct numeral recognition.The experimental results show that this App digital recognition system is convenient to use and achieves 94.77%, 98.32%, and 99.78% accuracy for handwriting numbers, print numbers, and ID numbers respectively.

关 键 词：深度学习 卷积神经网络 数字识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]