Abstract
WiFi channel state information (CSI) has emerged as a promising modality for human action recognition due to its non-invasive nature and robustness in diverse environments. However, most existing methods process CSI channels collectively, potentially overlooking valuable channel-specific information. In this study, we propose a novel architecture, DF-CNN, which treats CSI channels separately and integrates their outputs using a decision fusion (DF) strategy. Extensive experiments demonstrate that DF-CNN significantly outperforms traditional approaches, achieving state-of-the-art performance. We also provide a comprehensive analysis of individual and combined CSI channel evaluations, showcasing the effectiveness of our method. This work establishes the importance of separate channel processing in CSI-based human action recognition and sets a new benchmark for the field.