作者：刘思洋1, 黄坚1, 刘桂雄1, 罗文佳2
作者单位：1. 华南理工大学机械与汽车工程学院, 广东 广州 510640;
2. 广州市银科电子有限公司, 广东 广州 510663
关键词：法定货币;图像识别;深度学习;AlexNet
摘要：
基于法定货币在不同光照下局部特征不同，该文研究一种基于AlexNet的两光照下多类别法定货币识别技术。首先，分析自然光照、紫外光照下法定货币图像特征，指出不同光照下法定货币呈现不同的面额、图案等特征；其次，分析AlexNet神经网络模型与研究面向法定货币识别的AlexNet迁移学习方法；最后，在30类别的两光照下不同币种的图像样本库上进行图像识别实验，货币图像识别准确率达到100%，准确实现区分货币币种、光照条件、面额与正反面货币图像功能。与经典货币图像识别方法相比，该法能减少人工提取图像特征的工作量，具有通用性好、准确度高的特点。
Technology of multi-category legal currency identification under multi-light conditions based on AlexNet
LIU Siyang1, HUANG Jian1, LIU Guixiong1, LUO Wenjia2
1. School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510640, China;
2. Guangzhou Yin Ke Electronics Co., Ltd., Guangzhou 510663, China
Abstract: Based on the difference of local characteristics of legal currency under different illuminations, this paper studies a multi-category legal currency recognition technology based on AlexNet. Firstly, the characteristics of legal currency images under natural light and ultraviolet light are analyzed. It is pointed out that the legal currency presents different fetures and patterns under different illuminations. Secondly, the AlexNet neural network model and the AlexNet migration learning method for legal currency identification are analyzed. Finally, On the image sample library of different currencies under 30 kinds of illumination, the image recognition experiment is carried out, and the accuracy of currency image recognition reaches 100%, which accurately realizes the functions of distinguishing the kinds of currency, lighting conditions, denomination and front and back currency images. Compared with the currency image recognition method, the workload of manually extracting image features can be reduced, and the utility model has the advantages of good versatility and high accuracy.
Keywords: legal currency;image identification;deep learning;AlexNet
2019, 45(9):118-122 收稿日期: 2018-08-22;收到修改稿日期: 2018-09-29
基金项目: 广州市科技计划项目（2018020300006）
作者简介: 刘思洋(1995-),男,广东揭阳市人,硕士研究生,专业方向为精密检测与仪器仪表
参考文献
[1] 宋晓宁, 刘梓, 於东军, 等. 表格型票据图像手写体特殊符号的混合检测算法[J]. 南京理工大学学报(自然科学版), 2012, 36(6):909-914
[2] HASANUZZAMAN F M, YANG X, TIAN Y L. Robust and effective component-based banknote recognition for the blind[J]. IEEE Transactions on Systems Man & Cybernetics Part C, 2012, 42(6):1021-1030
[3] YONG K S, DANH P T, RYOUNG P K, et al. Recognition of banknote fitness based on a fuzzy system using visible light reflection and near-infrared light transmission images[J]. Sensors, 2016, 16(6):863
[4] 罗帅, 娄震. 基于印刷年份的人民币版本识别技术研究[J]. 现代电子技术, 2015, 38(18):72-74
[5] 金长江, 师廷伟. 红外弱小目标检测背景抑制算法研究[J]. 中国测试, 2016, 42(4):115-119
[6] 李雪梨, 索双富, 武佩君. 一种基于紫外荧光图像的人民币面额识别算法[J]. 机械设计与制造, 2017(3):1-3
[7] 郭雪梅, 刘桂雄. 多颜色模型分割自学习k-NN设备状态识别方法[J]. 中国测试, 2016, 42(4):107-110
[8] 叶龙欢, 王俊峰, 高琳, 等. 复杂背景下的票据字符分割方法[J]. 计算机应用, 2012, 32(11):3198-3200
[9] SOSA-GARCÍA J, ODONE F. Banknote recognition as a CBIR problem[C]//International Conference on Similarity Search and Applications. Springer, 2015.
[10] 韩梦迪, 曹玉东, 杜刚. 基于BP神经网络的银行票据识别[J]. 信息通信, 2016(9):157-159
[11] PHAM T D, NGUYEN D T, KIM W, et al. Deep learning-based banknote fitness classification using the reflection images by a visible-light one-dimensional line image sensor[J]. Sensors, 2018, 18(2):472
[12] 卢宏涛, 张秦川. 深度卷积神经网络在计算机视觉中的应用研究综述[J]. 数据采集与处理, 2016, 31(1):1-17
[13] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[C]//International Conference on Neural Information Processing Systems. 2012:1097-1105.
[14] LONG M S, WANG J M, DING G G, et al. Transfer feature learning with joint distribution adaptation[C]//IEEE International Conference on Computer Vision. IEEE, 2014:2200-2207.
