基于Wi-Fi信号特征增强的跨域人体行为感知方法OA

Wi-Fi devices are easy to deploy and widely available,making Wi-Fi signals an attractive me-dium for wireless sensing.In particular,Wi-Fi-based human activity recognition has shown great poten-tial in smart homes and human-computer interaction applications.However,existing approaches usually treat channel state information(CSI)merely as a time series,neglecting variations across subcarriers,and often suffer from poor generalization in cross-domain scenarios.To address these issues,this paper proposes a cross-domain behavior recognition framework based on adaptive multi-dimensional feature en-hancement with domain feedback,termed DFAE-Fi.The proposed framework jointly models the temporal and channel characteristics of CSI through a time-frequency feature encoder,and introduces a domain feedback-driven feature enhancement mechanism that adaptively adjusts attention weights according to CSI statistical properties and domain discrepancies.By establishing feedback connections from the do-main discriminator to the feature extractor,DFAE-Fi enables collaborative optimization of feature en-hancement and domain adaptation.In addition,a discriminative feature loss and a dual-path training strategy are incorporated to further improve feature separability.Experimental results on public datasets demonstrate that the proposed method consistently outperforms existing approaches across various cross-domain scenarios.