An efficient cross-domain device-free gesture recognition method for ISAC with federated transfer learning

Emerging device-free sensing technologies and applications promote the development of indoor ubiquitous sensing. Device-free sensing with machine learning mechanisms enable detection, recognition automatically, without required explicit programming. Because of concern of indoor sensing privacy and ubiquitous sensing ability, it is really necessary to conduct an in-depth survey on device-free sensing security training and cross-domain sensing issues. Existing surveys have two important problems: weak robustness and low efficiency. To address them, this article put forward to learn domain independent features, model training and inference localization based on federated transfer learning. Moreover, several efficient methods are proposed to provide a distributed edge device-free sensing mechanism with sensing data privacy protection, low time cost, communication, computing and energy resources. We implement the proposed mechanism and carry out experiments with Widar3.0 datasets to evaluate its performance. The results demonstrate that our mechanism performs better for cross-domain device-free sensing while preserving user data privacy and saving resources.