Multi-Grained Deep Cascade Learning for ECG Biometric Recognition

Recently, electrocardiogram (ECG) biometric recognition has become an emerging area of interest in the identity identification technology.However, robust and accurate of ECG biometric recognition are challenging.Sparse representation-based classification (SRC) and deep neural networks (DNNs) have achieved significant success in biometric recognition, but there are some problems, and SRC-based learning methods are one-step models where the latent discriminative information cannot be fully exploited.DNNs are complex models and require large amounts of training data.The proposed method describes the design of a multi-grained deep cascade learning for ECG biometric recognition that addresses above problems.First, the global and local features are generated by principal components analysis (PCA) and the use of one-dimensional multi-resolution local binary pattern (1DMRLBP) method.Second, we obtain new features of class coding based on the sparse representation by multi-granularity scanning.Third, we propose an end-to-end deep cascade learning model without back-propagation to seek more discriminative information.This approach not only effectively solves the one-step model problem of sparse representation, but also effectively reduces the ECG signal noise.Extensive experiments are conducted on four ECG datasets.The results demonstrate that the proposed method can outperform other state-ofthe-art methods in terms of both accuracy and efficiency.