Enhanced Human Activity Recognition Using Wi-Fi Sensing: Leveraging Phase and Amplitude with Attention Mechanisms

Wi-Fi-based human activity recognition (HAR) is a non-intrusive and privacy-preserving method that leverages Channel State Information (CSI) for identifying human activities. However, existing approaches often struggle with robust feature extraction, especially in dynamic and multi-environment scenarios, and fail to effectively integrate amplitude and phase features of CSI. This study proposes a novel model, the Phase-Amplitude Channel State Information Network (PA-CSI), to address these challenges. The model introduces two key innovations: (1) a dual-feature approach combining amplitude and phase features for enhanced robustness, and (2) an attention-enhanced feature fusion mechanism incorporating multi-scale convolutional layers and Gated Residual Networks (GRN) to optimize feature extraction. Experimental results demonstrate that the proposed model achieves state-of-the-art performance on three datasets, including StanWiFi (99.9%), MultiEnv (98.0%), and the MINE lab dataset (99.9%). These findings underscore the potential of the PA-CSI model to advance Wi-Fi-based HAR in real-world applications.