卷积神经网络的民国纸币序列号识别系统  

Paper currency serial number recognition system research of the Republic of China based on convolutional neural network

作　　者：沈成龙 王笑梅[1] 王晨[1] SHEN Chenglong;WANG Xiaomei;WANG Chen(College of Information,Electrical and Mechanical Engineering,Shanghai Normal University,Shanghai 201418,China)

机构地区：[1]上海师范大学信息与机电工程学院,上海201418

出　　处：《上海师范大学学报（自然科学版）》2020年第4期465-471,共7页Journal of Shanghai Normal University(Natural Sciences)

摘　　要：实现了深度学习的民国纸币序列号自动识别系统.提取、分割民国纸币序列号字符,对单个字符进行预处理,裁剪字符空白区域,归一化字符大小,并使用卷积神经网络进行识别.实验结果表明:在纸币存在污迹、褶皱的情况下,所提民国纸币序列号识别系统能够减少人工录入的工作量,单个字符的识别精度高于99.99%.An automatic recognition system of paper currency serial numbers of the Republic of China was realized by deep learning in this paper.Firstly,the characters of paper currency serial numbers of the Republic of China were extracted and segmented.Secondly,pre-processing for each character was conducted and the blank character zone was clipped in order to normalize the character size.Lastly,the characters were recognized by the convolutional neural network.The experimental results showed that the paper currency serial number recognition system proposed in the paper could reduce the workload of manual entry while there were stains and wrinkles existing on the paper currency.The recognition accuracy of a single character could reach more than 99.99%.

关 键 词：民国纸币 图像处理 序列号识别 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]