Research On Human Identification Method Based On Feature Extraction Of Pulse Signal

The human identification methods based on biometric features such as fingerprint or face have the defect of easy to be reproduced and forged because of non-living detection,while physiological signals are usually non-replicable and secure.Therefore,the human identification based on physiological signals is a good choice.Nowadays toward this direction the researchers mainly focused on study for ECG and heart sound signal.But,due to the installation location are private and inconvenient of the two kinds of signal sensors',they are probably not going to be popular.While PPG infrared pulse sensor is easy to use and to be integrated into system,so the human identification based on pulse signal features extraction was proposed in this thesis.The main research are as follows:(1)Study on pulse signal preprocessing and feature extraction methods.Referring to some classical methods of speech signal processing,the pulse signal is processed by bandpass filtering,normalization,framing and windowing.Average magnitude difference function(AMDF)based to calculate the pulse cycle.The time-domain features such as pulse rate feature,extremum feature,first order difference coefficient,second order difference coefficient and energy feature are extracted,and the non-uniform sub-band spectrum features are extracted based on down-sampling and short-time Fourier transform(STFT).(2)Study on identity-oriented pulse signal classification methods.In this thesis,dynamic time warping method(DTW)is used to classify feature samples of different lengths,and the others are classified by the statistical based recognition methods.The identification method based on DTW:the first order difference coefficients and the second order difference coefficients are used as features,and DTW is used to calculate the center of feature samples,and the minimum distance method based on DTW is used for classification.The recognition method based on statistical model:pulse rate feature,extremum feature,energy feature and sub-band spectrum features are respectively modeled by single Gaussian model and Gaussian Mixture Model,and the maximum likelihood method is used for recognition of statistical pattern.The experimental results of 46 subjects with the sampling rate of 1024Hz shows that:l)The highest average recognition rate of DTW recognition method is 53.54%.2)When the single Gaussian probability modeling and the maximum likelihood classification method are used,the recognition rate is the highest of 74.19%.The average recognition rate of the six subjects considering the scenario of family can be up to 99.12%.Besides,through the experimental results' comparison between sampling rate of 1024hz and 20kHz,it can be found that the recognition rate is not significantly improved when increasing the sampling rate significantly.(3)Study on pulse signal authentication method based on SVM.In this thesis,binary classifier of Support Vector Machines is used to classify pulse signals.Pulse rate feature,extremum feature,energy feature and sub-band spectrum features are connected in order as the input of SVM.Eight subjects were tested,and the results show that the average recognition rate can reach 81.04%by the SVM based method proposed.It can be seen from the experiment results that the method proposed in this thesis has certain practical value and the project is worthy of further study.Figure[62]Table[32]Reference[83]...