Hiding Identity, Preserving Respiration: Semantic-Decoupled Privacy Protection for mmWave Radar Sensing

Millimeter-wave (mmWave) radar-based respiratory sensing has attracted increasing attention due to its contactless, unobtrusive nature and high sensitivity to micro-motions. However, mmWave radar signals inevitably encode identity-sensitive information beyond respiration, raising serious privacy concerns. Existing privacy protection strategies are often coarsegrained or indiscriminately applied, leading to substantial degradation in respiratory sensing performance. To address these issues, we propose P 2 CRM , an identity-preserving mmWave radar-based contactless respiratory sensing system that explicitly balances sensing performance and identity protection. Specifically, by disentangling respiration rhythm from rhythm-irrelevant components, we design a Wavelet Packet Decomposition-based Semantic Decoupling ( WSD ) method to transform radar signals into a structured semantic representation. Building on this representation, a Key-Controlled Semantic Mixing (KCSM) mechanism is introduced to selectively perturb rhythm-irrelevant semantics to suppress identity cues while preserving respiration-critical information. While such semantic-level perturbation reduces identity leakage, it inevitably introduces distortions that impair respiratory sensing accuracy. To compensate for this perturbation-induced degradation, we further propose a Cross-Security Respiratory Knowledge Transfer (CRKT) network that transfers respiration-relevant knowledge from semantically rich signals to their perturbed representations, enabling accurate respiratory sensing directly from perturbed signals. Extensive experiments show that P 2 CRM reduces identity recognition accuracy from 83.38% to 28.47%, while maintaining a respiratory sensing MAE of 1.08 bpm. These results demonstrate that P 2 CRM achieves a favorable balance between reduced identity leakage while preserving respiratory sensing performance, enabling trustworthy and identity-preserving contactless respiratory monitoring. Our code is available at: https://github.com/xiaozhu990724-lxy/Trusted-BR-monitoring-P2CRM-.