ECG Based Biometric for Human Identification using Convolutional Neural Network

As growing falsification techniques threaten human identification, security is a major concern. ECG signals are irreplaceable and cannot be copied, so it is used for biometric application. In this paper, deep learning based one-dimensional convolutional neural network (1D-CNN) is proposed to classify ECG signals for biometric human identification. The proposed system focuses on the identification of ECG signals with lead I, lead II and combined lead I and lead II configurations. 10 layer CNN model is developed with ReLU activation function which increases the performance of the system. Wavelet transform (WT) approach is considered after segmentation of raw ECG signals. R-peak amplitude representing one heartbeat along with P and T wave amplitude values are given to CNN as a 1D matrix for identification. MIT-BIH (Normal Sinus Rhythm) and ECG-ID database are used for experimentation. With MIT-BIH (NSR) and ECG-ID database, an accuracy of 96.93% and 100% is achieved. It is observed that the Lead II configuration gives the highest accuracy as compared to lead I and combined configurations.