An Identity Recognition Technology Based On Photoplethysmography Signal

With the rapid development of information security technology, the traditional information security measures, such as complex digital passwords, personal identification documents, etc. can not meet people’s efficient, safe and convenient requirements. Therefore, currently fingerprint, iris and other biometric systems represented are widely used in financial transactions, computer networks and other applications, which greatly ease the urgent market need for information security protection. However, there are still some drawbacks in these biometric systems, such as easily being copied or forged. Hence, it is urgent to find new biological features to make up for these deficiencies, or take place of existing biological features for new biometric systems. This thesis selects the PPG signal as the research object, and does some research on two main identification methods based on the PPG signal:First, study an identification method based on P-wave fiducial points of the PPG signal. This method completes the original PPG signal preprocessing by the wavelet transform, and extracts the P-waves’ location of PPG signals by maximum-minimum method. Then, four-dimensional time-domain characteristics such as P-wave intervals, the maximum value of the P wave,etc. based on the P-waves’ location, are extracted as the training and test data for KNN classifier. Experimental results show that, under the conditions of large-scale training data, the correct recognition rate of the biometric system can achieve 85.34%.Second, study an identification method based on no-fiducial points of the PPG signal. This method firstly extracts a single period of PPG based on the extracted P-wave as a reference point for signal segmentation, interpolation, and then use the three data dimensionality tools, extended non-negative matrix factorization with sparse constraint, principal component analysis and nuclear component principal component analysis for data dimensionality reduction to extract features. Finally, the extracted features are input in both k-nearest neighbor and support vector machine for human identification. Comparing with the former method, the second method has the following advantages: PPG signal waveform characteristic is of anti-noise, high stability, distinctiveness. Evenin small training sample conditions, the identification method based on the PPG signal waveform feature also has a high rate of identification. The experimental results show that this method enjoys the correct recognition rate of 98.4%.