کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Self-Supervised Autoencoder-Based Suppression of Random Body Movements in Contactless Respiration Signals

نویسندگان :

Anahita Ghasemi¹ Mohammad Ali Sebt²

1- دانشگاه صنعتی خواجه نصیرالدین طوسی 2- دانشگاه صنعتی خواجه نصیرالدین طوسی

کلمات کلیدی :

contactless respiratory monitoring،signal denoising،self-supervised learning،autoencoder،random body movement

چکیده :

Contactless respiratory monitoring based on electromagnetic sensing signals has attracted significant attention in biomedical applications. Despite its advantages, respiration measurements obtained in this manner are highly susceptible to interference caused by random body movement (RBM), which can substantially degrade signal quality. Conventional signal processing techniques such as short-time Fourier transform (STFT) and nonnegative matrix factorization (NMF) are able to suppress motion-related artifacts; however, their high computational complexity limits their applicability in real-time scenarios. In this paper, a self-supervised autoencoder is proposed to effectively mitigate the impact of RBM in contactless respiratory signals. The autoencoder is trained using clean reference signals generated via STFT and NMF, allowing the model to learn the intrinsic mapping between motion-contaminated measurements and clean respiratory components. After training, the proposed method can automatically denoise unseen signals without the need for explicit signal decomposition. Experimental results obtained from real measurement data demonstrate notable improvements in signal quality and robustness, confirming the practicality and effectiveness of the proposed approach for real-time contactless respiratory monitoring.