Cross-Environment WiFi CSI Gesture Recognition Using a Domain-Generalized Multi-Scale Attention Residual Network with Robust Feature Learning

Abstract WiFi Channel State Information (CSI) records changes in wireless signals induced by hand and body motions, providing a practical and contact-free method of gesture recognition. Changes in the environment, multipath propagation, and signal circumstances might affect CSI patterns, making it tough to achieve consistent detection across varied indoor situations. In this paper, we provide a framework for WiFi CSI gesture recognition that works in any environment by using a robust feature-learning Domain-Generalized Multi-Scale Attention Residual Network. Meaningful patterns linked with various movements are identified by the framework via the use of CSI amplitude fluctuations throughout time and subcarriers. A Stable Field Various resolutions of the gesture are captured using multi-scale convolution, and the network is trained to concentrate on the most important features with the aid of channel and spatial attention. Stable model training and successful feature learning are both supported by residual connections. Evaluating performance in both learned and unknown situations, the trials take five gesture classes into account. The validation and training results for a robust domain demonstrate consistent recognition performance and continuous convergence. A Stable Field The framework’s recognition accuracy in the unseen test environment is 86%, while in the two training domains they are 94% and 91%, respectively. The t-SNE image further highlights the substantial differences across the gesture classes, suggesting that the learnt characteristics maintain their uniqueness even when exposed to different environments. A Stable Field According to the ablation research, each component contributes to the overall improvement in accuracy; for example, using multi-scale convolution increased accuracy from 72% to 79%, channel/spatial attention increased accuracy to 82%, and the full framework improved accuracy to 86%. A Stable Field The findings show that the suggested method can reliably and robustly recognize gestures based on WiFi CSI even when the interior environment is constantly changing.