A Self-attention Spiking Neural Network for Human-to-Human Interaction Recognition Based on Wi-Fi Signals

Leveraging Wi-Fi signals for human activity recognition (HAR) enables efficient deployment, cost savings, and enhanced privacy protection. Channel State Information (CSI) has become central in recognizing human-to-human interactions (HHIs) within HAR research. Traditional Artificial Neural Network (ANN) models, although widely used for HHIs recognition, suffer from high system overhead and suboptimal recognition accuracy. To address these limitations, we introduce an innovative Self-Attention Spiking Neural Network (SASNN) model, named HHIs-SASNN, specifically designed for recognizing HHIs. This model leverages the robust spatio-temporal processing capabilities of Spiking Neural Networks (SNN), augmented with a self-attention mechanism, to effectively extract and process spatio-temporal information from Wi-Fi signals. Our model exhibits superior performance in both accuracy and computational efficiency compared to traditional ANN-based methods. Our experimental results show that the proposed approach achieves an average accuracy of over 96% in recognizing 13 distinct types of human-to-human interaction behaviors, including steady-state activities, nearly a 1% improvement over conventional ANN methods. Additionally, the experiments indicate that our model achieves lower system overhead.