Research On ECG Signal For Human Identification Based On Deterministic Learning Theory

In today's society,with the rapid development of social economy and scientific information,traditional authentication methods cannot meet the growing demand for human security.Although many biometric recognition technologies(such as fingerprint,voice,face recognition and so on)have been widely used in security monitoring,financial security,medical and other fields,there also exists potential security risks of forgery.In this paper,a dynamic biometric recognition technology-ECG signal is studied.Its advantages such as vividness and high anti-counterfeiting have attracted wide attention in the field of biometrics recognition.Therefore,the application of ECG signal in identification can be a good complement to the existing biometrics recognition technology.Over the past decade,Wang et al.have carried out many innovative researches in the field of dynamic pattern recognition,and put forward a new machine learning theorydeterministic learning theory.It is a theory that uses RBF neural network to acquire,express,store and reuse knowledge in a dynamic unknown environment.It can realize local accurate identification of dynamic systems of periodic,periodic and even chaotic trajectories,and store learned dynamic knowledge in time-invariant RBF Neural network weights.Based on the deterministic learning mechanism,the paper proposes a dynamic ECG identification framework.The contributions of this paper lie in the following aspects:1)Acquisition and pre-processing of ECG data.On the basis of AIKD acquisition box,this paper compiles a PC program to collect standard 12-lead ECG signals,and studies a median filtering and wavelet filtering method for original ECG signals,which can effectively remove interference noise and provide an important guarantee for subsequent recognition.2)Extraction of cardiac dynamics features and identification.The time and frequency domain characteristics of ECG signals are limited and do not fully reflect the dynamic characteristics of ECG mode.This paper first converts standard 12-lead ECG data into 3-dimensional VCG data,and the conversion between them does not lose the information content related to cardiac dynamics.In the training phase,cardiac dynamics within ECG signals is extracted accurately by using RBF neural networks through deterministic learning mechanism.The cardiac dynamics of the signal are stored with time-invariant RBF neural network weights,and the extracted kinetic features are used to construct an estimator group to represent the trained ECG pattern.In the recognition phase,by comparing the test pattern with all the patterns of the dynamic estimator,a set of norm-form residuals can be obtained to measure the similarity between the recognition pattern and the training pattern.According to the principle of minimum residuals,the test pattern can be quickly identified.Finally,this paper carries out a series of experiments on ECG data collected by ourselves and ECG data from PTB database,and uses four evaluation indicators to verify the effectiveness and feasibility of this method.3)The ECG identification system based on MATLAB GUI is designed and implemented.It mainly includes three parts: data acquisition,ECG pattern training and ECG recognition,which provides an effective tool platform for the research and experimental analysis of this paper.