Dual-feature and multi-scale fusion using U-net deep learning model for ECG biometric recognition

Aiming at the challenges that the traditional photoplethysmography (PPG) biometrics is not robust and precision of recognition, this paper proposes a dual-feature and multi-scale fusion using U2-net deep learning model (DMFUDM). First, to obtain complementary information of different features, we extract the local and global features of one-dimensional multi-resolution local binary patterns (1DMRLBP) and multi-scale differential feature (MSDF). Then, to extract robust discriminant feature information from the 1DMRLBP and MSDF features, a novel two-branch U2-net framework is constructed. In addition, a multi-scale extraction module is designed to capture the transition information. It consists of multiple convolution layers with different receptive fields for capturing multi-scale transition information. At last, a two-level attention module is used to adaptively capture valuable information for ECG biometrics. DMFUDM can obtain the average subject recognition rates of 99.76%, 98.31%, 98.97% and 98.87% on four databases, respectively, and experiment results show that it performs competitively with state-of-the-art methods on all four databases.