ECG for Biometric Recognition Based on Hybrid Deep Neural Networks

This paper studies a popular biometrics technology, the application of the physiological characteristics of electrocardiogram (ECG) signals for human identification. The ECG signal is an effective and potential identification tool that can distinguish the typical characteristics of different people because it is universal in the human body and it is difficult to be forged. Based on the MIT-BIT database of normal sinus rhythm, this paper proposes a hybrid deep neural network model for biometric recognition using convolutional neural networks (CNN) and bi-directional long short-term memory (BiLSTM). The proposed model optimizes the features for classification by superimposing CNN onto the BiLSTM network layer, and achieves the effective mining and capture of the internal features of ECG signals, which to some extent overcomes the limitations of the convolution feature extraction methods. Furthermore, the performance comparison of different CNN-BiLSTM hybrid models (e.g. LeNet-BiLSTM and Inception-BiLSTM) is examined. The experimental results show that the Inception-BiLSTM hybrid model can produce a better result, and the identification accuracy is as high as 98%.