Squeeze-and-Excitation Hybrid Network for Biometric Identification via Photoplethysmography

Biometric identity verification is critical in security, healthcare, and personalized services. Photoplethysmography (PPG) signals, measuring blood volume changes in small vessels, offer a promising, non-invasive biometric method but often need more accuracy and robustness. To address this, we propose the ResNet-BiGRU-SE framework, which enhances PPG-based biometric identification through a s queeze-and-excitation (SE) attention mechanism. This architecture integrates convolutional neural networks (CNNs), bidirectional gated recurrent units (BiGRUs), and SE attention modules to extract spatial and temporal features from PPG signals. ResNet architecture in CNN layers extracts local patterns and mitigates vanishing gradients with residual connections, while BiGRU layers capture temporal dependencies and contextual information. The SE attention mechanism enhances focus on informative PPG signal parts by adjusting feature responses per channel. Evaluated on a standardized dataset of PPG signals from various patients, the ResNet-BiGRU-SE model demonstrates superior accuracy (95.26%), low loss (0.24), and a high F1-score (95.25 %), outper-forming recent methods. Comprehensive ablation studies confirm the model's robust and reliable performance.