一种基于卷积神经网络的纸币脏污识别方法  

Banknote Dirty Identification Method Based on Convolutional Neural Network

作　　者：孙伟忠 马跃[1,3] 尹震宇 谷艾[1,3] 徐福龙 SUN Wei-zhong;MA Yue;YIN Zhen-yu;GU Ai;XU Fu-long(University of Chinese Academy of Sciences,Beijing 100049,China;University of Science and Technology Liaoning,Anshan 114051,China;Shenyang Institute of Computing Technology,Chinese Academy of Science,Shenyang 110168,China)

机构地区：[1]中国科学院大学,北京100049 [2]辽宁科技大学计算机与软件工程学院,辽宁鞍山114051 [3]中国科学院沈阳计算技术研究所,沈阳110168

出　　处：《小型微型计算机系统》2020年第7期1508-1512,共5页Journal of Chinese Computer Systems

基　　金：辽宁省科学技术计划项目(2019JH2/10100019)资助;辽宁省高等学校首批重大科技平台开放课题(USTLGXZD201902)资助。

摘　　要：纸币的污损程度在某种程度上决定了纸币是否能够继续流通.如何精准的识别纸币的脏污,是当前金融机具面临的一项重大问题.为了解决这个问题,本文使用接触式图像传感器采集纸币在红光、绿光、蓝光、红外光下的双面反射图像,同时也采集纸币在绿光透射和红外光透射下的图像.通过使用图像处理的方法把纸币图像提取出来,然后分析不同脏污等级的纸币在各种光源照射下所形成的图像,最终决定把哪种光源的纸币图像输入到卷积神经网络.之后,将已经分类好的训练样本和测试样本通过上述方式处理,会得到纸币图像的训练样本和测试样本.使用训练样本对本文设计的卷积神经网络进行训练,就会得到本文所需要的纸币脏污识别分类器.然后使用测试样本在这个分类器上进行测试,会得到训练的分类器的识别效果.测试结果表明本文所设计的卷积神经网络分类器对于识别纸币脏污的准确性非常高.The degree of fouling of banknotes determines to some extent whether the banknotes can continue to flow.How to accurately identify the dirt of banknotes is a major problem facing financial instruments.To solve this problem,this paper uses a contact image sensor to collect double-sided reflection images of banknotes under red,green,blue,and infrared light,and also collects images of banknotes under green light transmission and infrared light transmission.The image of the banknote is extracted by using image processing,and then the image formed by the banknotes of different soiling levels under various light sources is analyzed,and finally,which banknote image of the light source is input to the convolutional neural network.Thereafter,the trained training samples and test samples that have been classified are processed in the above manner,and training samples and test samples of the banknote image are obtained.Using the training samples to train the convolutional neural network designed in this paper,we will get the banknote dirty identification classifier needed in this paper.Then use the test sample to test on this classifier,you will get the recognition effect of the trained classifier.The test results show that the convolutional neural network classifier designed in this paper has very high accuracy for identifying banknotes.

关 键 词：纸币脏污识别 CNN 接触式图像传感器 

分 类 号：TP394[自动化与计算机技术—计算机应用技术]