Federated Learning-based Framework for Cross-Environment Human Action Recognition Using Wi-Fi Signal

Wi-Fi-based Human action recognition (HAR), as significant support for the loT applications, such as human-computer interaction, healthcare, etc. is attracting the attention of more and more researchers. With the rapid development of deep learning (DL), the DL-based HAR methods achieve excellent performance. Even though, the generalization performance of cross-environment HAR is still a challenge. Previous work relies on collecting sufficient data in different environments, which is time-consuming and labor-constraint. To address this problem, in this paper, we proposed a cloud-edge paradigm-based framework named WiFed-Sensing. In this framework, a personalized federated learning strategy is proposed to learn the general human action knowledge that cross-environment, which can make the HAR in new environments benefit from it and realize reliable HAR performance even with only a few action samples, thus improving the overall cross-environment HAR accuracy. Extensive experiments are conducted to evaluate the effectiveness of our framework, and the results demonstrate that our method achieves 89.52% cross-environment HAR accuracy, which outperforms the state-of-the-art method.