PlexNet: A fast and robust ECG biometric system for human recognition

Abstract

Researchers have explored the potential of electrocardiogram (ECG) to use as biometrics from past two decades. ECG has the inherent feature of vitality for securing the biometric system from fraudulent attacks. This paper proposes a novel ensemble of the state-of-the-art pre-trained deep neural networks i.e., ResNet and DenseNet for ECG biometric recognition. The principle of transfer learning is utilized to prepare fine-tuned models. The gathered knowledge of four fine-tuned models is fused to prepare one stacking model i.e., ‘PlexNet’. The PlexNet takes advantage of transfer learning along with ensemble learning, thus making a novel model for ECG biometrics that is robust and secure than other methods using deep networks. Two public datasets PTB and CYBHI are tested on the proposed ensemble for human identification. The experimental results demonstrate the efficacy of the model with identification accuracy reported the best as 99.66% on healthy and unhealthy subjects. Finally, the proposed ECG biometric method proves its robustness from signal acquisition methods, size of datasets, and subject health statuses.