PulsePrint: Single-arm-ECG biometric human identification using deep learning

Abstract

Focusing on the privacy and security challenges brought by emerging/promising smart health applications, we propose a single-arm-ECG biometric human identification system, with two major contributions. Firstly, to replace the traditional inconvenient/uncomfortable ECG leads like the chest and two-wrist lead configurations, we propose a highly wearable single-arm-ECG lead configuration. Secondly, to prevent time-consuming and information-missing feature engineering work, we introduce advanced deep learning techniques to automatically learn from the raw ECG data highly level features. To achieve this goal, the 1D ECG time series is transform to a new domain, where a 2D ECG representation is obtained. Afterwards, a convolutional neural network is applied to the 2D ECG data and learn the hidden patterns for user identification purpose. The proposed system is validated on a single-arm-ECG dataset. This study demonstrates the feasibility of this highly wearable deep learning-empowered human identification system.