Research On Identification Of ECG Signal Based On Convolutional Neural Network

With the rapid development of science and technology,people's requirements for information security are getting higher and higher.Due to the living characteristics of the ECG signal,the identity recognition based on the ECG signal can effectively avoid spoofing attacks.Generally,the identity recognition based on ECG signals needs to extract stable features from a period of time ECG signals to ensure high recognition accuracy.However,too long acquisition time will cause inconvenience to the user and reduce the convenience of the ECG signal identification system in practical applications.Aiming at how to improve the practicability of ECG signal identification,this article uses the deep learning method to study short-term ECG signal identification.The main research contents are as follows:To improve the recognition accuracy of short-term ECG identification,we propose one-dimensional convolutional neural network model based on residual block in this paper.By increasing the number of convolution layers and convolution cores,the network model can extract more sufficient features of ECG and improve the classification performance of the model.The shortcut connection design is used to solve the problem of performance degradation of deep convolutional neural network.The model only need a single cardiac cycle signal(less than 1 second)for identification,which greatly shortens the ECG acquisition time and improves the convenience of ECG identification.Experiments are performed on two public databases of ECG-ID and PTB.The experimental results show that the recognition accuracy of the model on ECG-ID and PTB test sets reaches 97.06% and 99.14% respectively.Compared with traditional machine learning methods such as SVM,the recognition accuracy of the model was improved by 7.16% and 1.46%,respectively.To build a network model with higher classification effect by using fewer sample sets,a method of ECG sample data enhancement based on Matthew effect is proposed in this paper.The method uses limited real samples to generate more virtual samples,which enriches the training sample data set.Compared with the recognition results of the model before sample enhancement,the recognition accuracy of the classification model on ECG-ID and PTB test sets was improved by 1.24% and 0.51% respectively.Experimental results shows that the proposed data enhancement method can improve the classification ability of the network model to a certain extent.