Research On Electrocardiogram Identificationwith Unknown Rejection And Multi-stage Neural Network

With the high speed development of the information age, human identification system is becoming more and more important for people’s daily li fe, such as in permission recognition of an access control system, confirmation management of financial transactions, and accounts management of electronic commerce and so on. As a matter of fact, most system need a mechanism for human identification system, so that the security service can be guaranteed. Compared to trad itional mechanism like password or command, the emergence of human identification system based on biometric feature is more security and reliable.Electrocardiogram(ECG) is the recording of the electrical activity of the heart and it can be detected by electrodes attached to the surface of the skin and recorded or displayed by a device external to the body. The advantage of using ECG for human identification system includes: persistence, uniqueness, highly attack resistance and so on.This thesis first proposes an unknown rejection algorithm based on neural network for human identification. This algorithm uses maximum value and second largest value of neural network output nodes to perform rejection policy with two thresholds. Then global shape feature extraction and local statistic feature extraction method are proposed. The global shape feature is basically normalization of an R peak to R peak interval of an ECG signal. The statistical feature represent the difference of amplitude of ECG signals. This thesis proposes a two-stage neural network classification structure based on unknown rejection algorithm to better utilize these two kind features. Experiment results show that the two-stage neural network structure is better than any single feature and als o better than linear splice of the two kind features. At last, this thesis proposes a multi-stage neural network classification structure based on unknown rejection algorithm. And the advantage of multi-stage classification has been discussed both on the research aspects and theory aspects. Two multi-features extraction methods are used. One is changing the single interval set to multi-interval sets for local statistical feature. The other is using wavelet decomposition for some raw continuous R peak to R peak intervals of ECG signals. Experiments results show that multi-stage neural network is better than single neural network on ECG identification system. And also experiment results show that appropriately increase the number of neural network stages can improve the effect of ECG identification system at a certain extent.