ECG-based Human Identification System by Adaptive SVD Algorithm

At present, people have higher requirements for the security of biometric technology. Therefore, ECG-based human identification is attracting more attention. In this paper, we designed a portable ECG acquisition device and proposed an adaptive SVD algorithm for ECG signal feature extraction in identity recognition. Compared with the classical singular value decomposition-total least square (SVD-TLS) algorithm, the adaptive SVD algorithm can automatically select the model and order of the parameterized power spectrum by the correlogram. It makes up for the defect of SVD-TLS algorithm that it isn’t easy to select an appropriate threshold in normalized ratio. In addition, due to the characteristics of ECG signals in frequency domain, this method can reduce the number of features and further increase the execution efficiency of identification network. The research of this paper is as follows: First, multiple cardiac cycles are segmented from the ECG signal. Second, the feature of each cardiac cycle is obtained by the adaptive SVD algorithm. Then, features are applied to the convolutional neural network (CNN) to realize the identification of ECG signals. Finally, we also developed ECG signal analysis software, which can obtain various ECG signal analysis results more conveniently. We tested the proposed method in MIT-BIH Arrhythmia database and the identification performance is reported to 99.59%. At the same time, ECG data of 40 healthy people were collected by the designed portable ECG acquisition device, and achieved a rate of 95.03% for the self-built database. Experimental results show that this method has better performance than the existing methods.