基于手指心电信号分析的个体身份辨识算法研究

发布时间：2018-06-07 02:51

本文选题：手指心电信号 + 生物识别　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：在高度信息化的现代社会中,随着交通、网络和通讯的飞速发展,人类的活动范围越来越大,在实现个体的身份识别的时候,人们面临的安全问题也越来越严重。生物特征识别技术(Biometrics)是运用个体一些独有的并且长期稳定的生物特征进行身份辨识的一种技术,例如虹膜、指纹、脸象和声音、笔迹、步态等,具有很高的准确度与可靠度,但也同样面临着安全问题,如假冒指纹、虚假虹膜、笔迹的模仿等,所以具备高防伪性的新颖生物特征识别技术迫切需要被提出。心电信号(Electrocardiogram,ECG)包含着人体独一无二的身份信息,突出特点是具有很高的防伪能力。而且近年随着小体积、低能耗、无需导电胶和易集成的心电采集芯片的出现,实现了在手指端实现心电的采集,可以普遍的应用到家庭生活当中,更方便快捷的实现个体身份识别。本文提出了基于手指心电信号分析的个体身份辨识算法的研究。首先,对手指心电信号用小波软阈值和遗传算法相结合的方法进行去噪;其次,研究了手指心电信号的稀疏特性,提出了基于KSVD+PCA下稀疏编码的手指心电信号身份识别算法和基于改进的标签一致LC-KSVD的手指心电身份识别算法,最后并采用两个手指心电数据库验证了这两个算法,取得了较大的识别率。本文的工作主要为:1、阐述了心电信号的产生机理、波形特点及ECG采集方式和基于手指心电的身份识别算法的发展,介绍了ECG的识别评价指标和两个手指心电数据库,为手指心电信号应用于个体身份识别技术领域提供了理论基础。2、提出基于K-SVD+PCA下稀疏编码的手指心电身份识别算法。首先对手指ECG用小波软阈值和遗传算法相结合的方法去噪,经过R峰检测、单周期划分、归一化,得到P-QRS-T单周期波群,结合手指心电特性,提取P-QRS-T波群构成特征向量并构建字典模板模型,用KSVD+PCA训练成冗余字典,然后对每一部分特征向量进行稀疏编码,实现在该字典上的稀疏表示。最后利用两个心电数据库(CYBHi,Surface ECG data)测试了算法性能,取得98.333%和100%的识别率。3、提出基于改进的标签一致LC-KSVD的手指心电身份识别算法。首先提取手指心电信号的平均单周期的P-QRS-T波群作为训练样本,然后提出自适应子字典和可调类标签对标签一致性的LC-KSVD(Label consistent KSVD,LC-KSVD)进行改进,然后利用改进的LC-KSVD1和LC-KSVD2算法完成识别。当输入信号和字典原子之间的标签信息相互之间一一对应的时候,在目标函数中,把判别误差、重构误差和分类误差结合起来一起用K-SVD算法来学习,更新字典和训练一个分类器。最后通过两个手指心电信号数据库(CYBHi,Surface ECG data)对本文的算法进行了性能测试,取得99%和100%的识别率。本文提出了基于手指心电信号分析的个体身份辨识算法研究,为手指ECG身份识别技术的实用化奠定了理论基础和技术支撑。
[Abstract]:In the highly information-based modern society, with the rapid development of traffic, network and communication, the range of human activities is becoming wider and larger, and the security problems that people face are becoming more and more serious when the identity of individuals is realized. Biometrics (Biometrics) is a technique for identity identification using individual biometric features that are unique and stable over a long period of time, such as iris, fingerprints, faces and sounds, handwriting, gait, etc., with high accuracy and reliability. However, it also faces security problems, such as fake fingerprints, false iris, imitation of handwriting and so on. Therefore, a novel biometric identification technology with high security needs to be put forward urgently. Electrocardiogramme (ECG) contains unique identity information of human body, which is characterized by its high anti-counterfeiting ability. And in recent years, with the emergence of small volume, low energy consumption, no conductive glue and easy integration of ECG acquisition chips, ECG acquisition at the finger end has been realized, which can be widely used in family life. More convenient and quick implementation of individual identity recognition. In this paper, an individual identification algorithm based on finger ECG signal analysis is proposed. Firstly, the wavelet soft threshold and genetic algorithm are used to Denoise the finger ECG signal. Secondly, the sparse characteristic of the finger ECG signal is studied. In this paper, an algorithm of finger ECG identification based on sparse coding under KSVD PCA and an algorithm based on improved label consistent LC-KSVD are proposed. Finally, two finger ECG databases are used to verify the two algorithms. A large recognition rate was obtained. The main work of this paper is 1: 1. The mechanism of ECG signal generation, the characteristics of waveform, the development of ECG acquisition method and the identification algorithm based on finger ECG are described. The recognition evaluation index of ECG and two finger ECG databases are introduced. This paper provides a theoretical basis for the application of finger electrocardiogram in the field of individual identity recognition, and proposes an algorithm of finger ECG identity recognition based on sparse coding under K-SVD PCA. First of all, the finger ECG is denoised by wavelet soft threshold and genetic algorithm. After R peak detection, single cycle partition, normalization, P-QRS-T single cycle wave group is obtained and combined with finger electrocardiogram. The P-QRS-T wave group is extracted to form the feature vector and the dictionary template model is constructed. Then the redundant dictionary is trained by KSVD PCA and then each part of the feature vector is sparse encoded to realize the sparse representation on the dictionary. Finally, the performance of the algorithm is tested by using two ECG databases (CYBHiPSurface ECG data), and the recognition rates of 98.333% and 100% are obtained. A finger ECG identification algorithm based on improved tagged consistent LC-KSVD is proposed. First, the average single-period P-QRS-T wave group of finger ECG is extracted as the training sample, then an adaptive sub-dictionary and a adjustable class label are proposed to improve the label consistency of LC-KSVD(Label consistent KSVDU LC-KSVD, and then the improved LC-KSVD1 and LC-KSVD2 algorithms are used to complete the recognition. When the label information between the input signal and the dictionary atom corresponds one by one, in the objective function, the discriminant error, the reconstruction error and the classification error are combined to learn by the K-SVD algorithm. Update the dictionary and train a classifier. Finally, the performance of the proposed algorithm is tested by using two finger ECG database, CYBHiP Surface ECG data, and the recognition rates of 99% and 100% are obtained. In this paper, an individual identity identification algorithm based on finger ECG signal analysis is proposed, which lays a theoretical foundation and technical support for the practical application of finger ECG identification technology.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN911.6

【参考文献】