【作者】 杨晓玲；

【导师】 同鸣； 沈阳；

【作者基本信息】 西安电子科技大学 ， 工程硕士（专业学位）， 2017， 硕士

【摘要】 在高速发展的现代化信息社会,大众信息的安全交流受到了越来越大的挑战,保证信息安全交流的技术的研究与开发也显得意义越来越重大。生物识别是一种通过人体生理的信号或者行为的独特特征自动进行个体身份确认的技术。传统的人脸、指纹、虹膜等认证方法虽广泛用于金融交易等领域,但基于这些特征的识别技术也出现了越来越多的问题,不能满足对信息安全要求越来越高的标准。心电信号、光电容积脉搏波信号以及呼吸信号等,作为人体固有的生理信号,可以反映独特的个体特征且极易获取,在身份识别领域有较大的应用价值及研究空间,本文以光电容积脉搏波和呼吸信号作为身份识别的研究对象,提出了两种身份识别算法:(一)提出了一种基于PPG信号最佳周期波形分段的识别方法。该方法首先以单周期波形的幅值和宽度作为选取准则,选取幅值和宽度在某一范围内的单周期波形,作为最佳周期波形;然后对单周期波形进行分段,并将每一段波形间的相似性作为该段的权重因子,利用KPCA方法对各段波形进行特征提取,并利用权重因子对各段波形特征向量加权融合,得到融合后的特征向量,利用支持向量机分类器进行分类识别。(二)提出了一种基于正则化广义局部判别典型相关分析的生理信号融合身份识别方法。该方法是在典型相关分析的基础上,通过添加正则化参数来消除噪声对特征提取的干扰,同时计算局部类内、类间相关系数矩阵,将样本的类别信息添加到特征提取之中,而且还计算了类内散度约束项和类间散度约束项使同类样本间的间距变小,异类样本间的间距增大,得到一种正则化广义局部判别典型相关分析算法,利用该算法对PPG信号和呼吸信号进行特征融合,提出一种基于正则化广义局部判别典型相关分析的生理信号融合身份识别方法。最终,实验仿真结果得出99.5%的身份识别率,证明了该方法的有效性。更多还原

【Abstract】 As the modern information society develops rapidly,the secure communication of public information has been greatly challenged.It is becoming more and more important to develop new technology or doing new researches to ensure secure communication.Biometric is a technology which can automatically identify the individual identity through the physiological signals or unique characteristics of the human behavior.The custom authentication methods which are based on faces,fingerprints or iris are widely used in financial transactions and other fields.However,there are more and more problems in these methods and they cannot meet the requirements of the situation which need high information security.As inherent physiological signals,ECG signal,Photoplethysmography signal and respiratory signal can reflect individual unique characteristics,and that they could be easily acquired.They have great application value and a wide research space in the field of identification.In this paper,two kinds of algorithms based on Photoplethysmography signal and respiratory signal about identification are proposed:(一)This paper presents a method of identification based on the segments of the optimal periodic waveform of PPG signal.Firstly,take the amplitude and width of the single period waveform as the selection criteria and the single-period waveforms whose amplitudes and widths within a certain range are chosen as the optimal waveforms.Then the one-cycle waveforms are divided into several parts.The similarity between the waveforms which are in the same segment is calculated as the weight.At the same time,features are extracted by using KPCA method.Using the weighting factors,the weighted fusion of the waveforms are conducted by fusing the feature vectors of every segment.Finally,the fused eigenvectors are predicted to which category by using the SVM classifier.(二)A physiological signal fusion identification method based on the regularized generalized local discriminant canonical correlation analysis is proposed.The method is based on the canonical correlation analysis.By adding a regularization parameters,the noise interference with feature extraction are eliminated,and the category information of the samples are added to the extracted feature by computing the local intra-classcorrelation matrix and the local inter-class correlation matrix.And also the intra-class divergence constraints are computed to make the smaller space between the similar samples and the inter-class divergence constraints are computed to make the larger space between the heterogeneous samples.Then a kind of regularization of generalizes local discriminant canonical correlation analysis algorithm is presented.By using the algorithm,the features of PPG signal and respiration signal are extracted and fused.Finally,the simulation results show the validity of the method by getting the 99.5% rate of identification.更多还原