采用轻量化神经网络的高安全手指静脉识别系统

李佳阳; 周颖玥; 杨阳; 李小霞

采用轻量化神经网络的高安全手指静脉识别系统

李佳阳¹,
周颖玥^{1, 2, ,},
杨阳¹,
李小霞^{1, 2}

1.
西南科技大学信息工程学院, 四川绵阳 621010
2.
西南科技大学特殊环境机器人技术四川省重点实验室, 四川绵阳 621010

基金项目:

四川省科技计划资助 2021YFG0383

西南科技大学龙山学术人才科研支持计划 17LZX648

西南科技大学龙山学术人才科研支持计划 18LZX611

详细信息

作者简介:
李佳阳(1997-)，男，硕士研究生，研究方向为图像处理、模式识别

通讯作者:
周颖玥(1983-)，女，副研究员，主要研究方向为计算机视觉、模式识别、医学图像处理。E-mail: zhouyingyue@swust.edu.cn

中图分类号: TP391.4
计量
- 文章访问数: 96
- HTML全文浏览量: 68
- PDF下载量: 37
出版历程
- 收稿日期: 2022-07-29
- 修回日期: 2022-09-17
- 刊出日期: 2024-02-19

High-Security Finger Vein Recognition System Using Lightweight Neural Network

LI Jiayang¹,
ZHOU Yingyue^{1, 2, ,},
YANG Yang¹,
LI Xiaoxia^{1, 2}

1.
School of Information Engineering, Southwest University of Science and Technology, Mianyang 621010, China
2.
Sichuan Provincial Key Laboratory of Robot Technology Used for Special Environment, Southwest University of Science and Technology, Mianyang 621010, China

摘要

摘要: 针对特殊材料能伪造手指静脉从而欺骗识别系统，以及利用卷积神经网络进行手指静脉识别计算量大的问题，设计了具有活体检测功能和轻量化卷积神经网络结构的手指静脉识别系统。采用光容积法检测手指脉搏波的变化，从而判断被采集对象是否为活体；利用剪枝及通道恢复方法改进了ResNet-18卷积神经网络，并结合L₁正则化增加卷积神经网络的特征选择能力，在提升算法准确率的基础上，能有效地降低计算资源的消耗。实验表明，使用改进的剪枝及通道恢复优化结构，参数量降低了75.6%，计算量降低了25.6%，在山东大学和香港理工大学手指静脉数据库上得到的等误率分别为0.025%、0.085%，远低于ResNet-18得到的等误率(0.117%、0.213%)。
- 手指静脉识别系统 /
- 活体检测 /
- 剪枝 /
- 通道恢复
Abstract: Special materials can fake finger veins to deceive recognition systems, and the large amount of computation is required for finger vein recognition using convolutional neural networks. In this study, a finger vein recognition system with an in vivo detection function and a lightweight convolutional neural network structure was designed. Changes in finger pulse waves were detected using Photo Plethysmo Graphy to determine whether the object was living. The ResNet-18 convolutional neural network was improved using a pruning and channel recovery method, and L₁ regularization was combined to improve the feature selection ability of the convolutional neural network. On the basis of improving the accuracy of the algorithm, the network can effectively reduce the consumption of computing resources. The experimental results show that by using the improved pruning and channel recovery optimization structure, the number of parameters decreases by 75.6%, the amount of calculations decreases by 25.6 %, and the iso-error rates obtained in the Shandong University and Hong Kong Polytechnic University finger vein databases are 0.025% and 0.085%, respectively, which are significantly lower than those of RESNET-18 (0.117% and 0.213%).
- finger vein recognition system /
- vivo detection /
- pruning /
- channel recovery

HTML全文

图 1 残差结构示意图

Figure 1. Schematic diagram of residual structure

下载: 全尺寸图片幻灯片

图 2 优化恒等结构

Figure 2. Optimized identity structure

下载: 全尺寸图片幻灯片

图 3 PRC Conv结构图

Figure 3. The structure of PRC Conv

下载: 全尺寸图片幻灯片

图 4 Pruning Conv流程

Figure 4. Flow chart of Pruning Conv

下载: 全尺寸图片幻灯片

图 5 通道恢复流程

Figure 5. Flow chart of channel recovery

下载: 全尺寸图片幻灯片

图 6 手指静脉识别系统结构

Figure 6. The structure of finger vein recognition system

下载: 全尺寸图片幻灯片

图 7 手指静脉图像采集部分结构：(a) 图像采集装置结构图；(b) 脉搏传感器；(c) 图像传感器

Figure 7. The structure of image acquisition device: (a) The structure of image (b) Pulse sensor; (c) Imaging sensor acquisition device

下载: 全尺寸图片幻灯片

图 8 SDUMLA和HKpolyU静脉示例图

Figure 8. Example diagram of veins in SDUMLA and HKpolyU

下载: 全尺寸图片幻灯片

图 9 不同结构提取的浅层特征

Figure 9. Shallow feature maps extracted from different structures

下载: 全尺寸图片幻灯片

图 10 L₁正则化对比

Figure 10. Comparison of L₁ regularization

下载: 全尺寸图片幻灯片

图 11 SDUMLA和HkpolyU数据集上EER值对比

Figure 11. Comparison of EER values on SDUMLA and HkpolyU datasets

下载: 全尺寸图片幻灯片

表 1 不同识别方法的比较

Table 1 Comparison of different recognition methods

Methods	Accuracy/%		Parameters	Flops
Methods	SDUMLA	HKpolyU	Parameters	Flops
AlexNet	96.32	92.28	26.7×10⁶	1.7×10⁸
Hong’s method^[23]	95.77	95.83	44.0×10⁶	32×10⁸
ResNet-18	97.79	96.26	19.7×10⁶	4.3×10⁸
PRC	98.46	98.28	4.8×10⁶	3.2×10⁸
PRC+L₁	99.42	98.86	4.8×10⁶	3.2×10⁸

下载: 导出CSV

参考文献(23)

[1]	Jain A, Bolle R, Pankanti S. Biometrics: Personal Identification in Networked Sciety[M]. Springer Science & Business Media, 1999.
[2]	黄易豪. 基于手部静脉图像的身份识别关键技术研究[D]. 绵阳: 西南科技大学, 2021. HUANG Yihao. Research on Key Technologies of Identity Recognition Based on Hand Vein Images[D]. Mianyang: Southwest University of Science and Technology, 2021.
[3]	Mohsin A H, Zaidan A A, Zaidan B B, et al. Finger vein biometrics: Taxonomy analysis, open challenges, future directions, and recommended solution for decentralised network architectures[J]. IEEE Access, 2020, 8: 9821-9845. DOI: 10.1109/ACCESS.2020.2964788
[4]	Ahmad Radzi S, Khalil-Hani M, Bakhteri R. Finger-vein biometric identification using convolutional neural network[J]. Turkish Journal of Electrical Engineering and Computer Sciences, 2016, 24(3): 1863-1878.
[5]	Das R, Piciucco E, Maiorana E, et al. Convolutional neural network for finger-vein-based biometric identification[J]. IEEE Transactions on Information Forensics and Security, 2019, 14(2): 360-373. DOI: 10.1109/TIFS.2018.2850320
[6]	HONG H G, Lee M B, Park K R. Convolutional neural network-based finger-vein recognition using NIR image sensors[J]. Sensors, 2017, 17(6): 1297-1318. DOI: 10.3390/s17061297
[7]	LI R, SU Z, ZHANG H, et al. Application of improved GCNs in feature representation of finger-vein[J]. Journal of Signal Processing, 2020, 36(4): 550-561.
[8]	HE X, CHEN X. Finger vein recognition based on improved convolution neural network[J]. Comput. Eng. Design, 2019, 40: 562-566.
[9]	DAI Y, HUANG B, LI W, et al. A method for capturing the finger-vein image using nonuniform intensity infrared light[C]//2008 Congress on Image and Signal Processing, 2008: 501-505.
[10]	YU X, YANG W, LIAO Q, et al. A novel finger vein pattern extraction approach for near-infrared image[C]//2009 2nd International Congress on Image and Signal Processing, 2009: 1-5.
[11]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
[12]	陈鹤森. 基于深度学习的细粒度图像识别研究[D]. 北京: 北京邮电大学, 2018. CHEN Hesen. Research on Fine-Grained Image Recognition Based on Deep Learning[D]. Beijing: Beijing University of Posts and Tele-Communications, 2018.
[13]	LIU J, ZHUANG B, ZHUANG Z, et al. Discrimination-aware network pruning for deep model compression[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(8): 4035-4051.
[14]	HAN K, WANG Y, TIAN Q, et al. Ghostnet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 1580-1589.
[15]	弓雷. ARM嵌入式Linux系统开发详解[M]. 2版: 北京: 清华大学出版社, 2014. GONG Lei. ARM Embedded Linux System Development Details[M]. 2nd edition: Beijing: Tsinghua University Publishing House Co. Ltd., 2014.
[16]	CHEN X, HUANG M, FU Y. Simultaneous acquisition of near infrared image of hand vein and pulse for liveness dorsal hand vein identification[J]. Infrared Physics & Technology, 2021, 115: 103688.
[17]	WU W, YUAN W, LIN S, et al. Selection of typical wavelength for palm vein recognition[J]. Acta Optica Sinica, 2012, 32(12): 1211002. DOI: 10.3788/AOS201232.1211002
[18]	LIU J, YAN B P-Y, DAI W-X, et al. Multi-wavelength photo-plethysmography method for skin arterial pulse extraction[J]. Biomedical Optics Express, 2016, 7(10): 4313-4326. DOI: 10.1364/BOE.7.004313
[19]	YIN Y, LIU L, SUN X. SDUMLA-HMT: a multimodal biometric database[C]// Chinese Conference on Biometric Recognition(CCBR), 2011: 260-268.
[20]	Kumar A, ZHOU Y. Human identification using finger images[J]. IEEE Transactions on Image Processing, 2011, 21(4): 2228-2244.
[21]	罗建豪. 面向深度卷积神经网络的模型剪枝算法研究[D]. 南京: 南京大学, 2020. LUO Jianhao. Research on Model Pruning Algorithmsfor Deep Convolutional NeuralNetworks[D]. Nanjing: Nanjing University, 2020.
[22]	LI H, Kadav A, Durdanovic I, et al. Pruning filters for efficient convnets[J/OL]. arXiv preprint arXiv: 1608.08710, 2016.
[23]	HONG H G, Lee M B, Park K R. Convolutional neural network-based finger-vein recognition using NIR image sensors[J]. Sensors, 2017, 17(6): 1297. DOI: 10.3390/s17061297

施引文献(27)

期刊类型引用(12)

1.	刘和，宋璎珞，胡龙湘，刘国辉，王侃，王爱丽. 基于空间金字塔注意力机制残差网络的高光谱图像分类. 液晶与显示. 2024(06): 833-843 . 百度学术
2.	吴庆岗，刘中驰，贺梦坤. 融合MS3D-CNN和注意力机制的高光谱图像分类. 重庆理工大学学报(自然科学). 2023(02): 173-182 . 百度学术
3.	张云龙，齐国红，许新华. 基于模式识别技术的高光谱图像分类研究. 激光杂志. 2023(07): 95-99 . 百度学术
4.	贺敏慧，何敬，刘刚. 改进的混合2D-3D卷积神经网络高光谱图像分类研究. 时空信息学报. 2023(02): 184-192 . 百度学术
5.	张晓瑞. 基于卷积神经网络的多标签图像分类识别算法研究. 通化师范学院学报. 2022(02): 75-82 . 百度学术
6.	王欣，樊彦国. 基于改进DenseNet和空谱注意力机制的高光谱图像分类. 激光与光电子学进展. 2022(02): 211-222 . 百度学术
7.	孔祥魁，樊翠红. 多特征融合和最小二乘支持向量机的运动视频图像分类研究. 南京理工大学学报. 2022(02): 164-169+176 . 百度学术
8.	郑爽，梁云浩，武俊峰，乔壮，刘付刚. 基于AlexNet-SN网络的煤与煤矸石分类方法. 中国矿业. 2022(06): 79-85 . 百度学术
9.	王晨海，陈澳，王康乐. 虚拟现实环境下基于红外光谱的多特征纹理图像视觉传达方法研究. 激光杂志. 2022(11): 154-158 . 百度学术
10.	吴鸿昊，王立国，石瑶. 高光谱图像小样本分类的卷积神经网络方法. 中国图象图形学报. 2021(08): 2009-2020 . 百度学术
11.	吴涛. 基于特征提取和半监督学习的图像分类算法. 粘接. 2021(11): 92-97 . 百度学术
12.	杨远陶，刘瑞，曹礼刚，杨梅，陈景珏. 基于随机森林算法的珠海一号高光谱影像土地利用信息提取. 物探化探计算技术. 2021(06): 818-824 . 百度学术