Scholarlink

<p>This paper introduces a novel non-contact physiological signal detection method using Ultra-Wideband (UWB) radar, integrating Kalman tracking and time-frequency deep learning to mitigate range gate jitter and noise in multi-target scenarios. A dynamic threshold adaptive peak detection method, combined with multi-Kalman filtering, ensures robust target acquisition. A deep learning model with frequency domain enhancement improves respiratory rate (RR) and heart rate (HR) detection accuracy. The framework employs multi-task learning, optimizing RR regression, HR classification, enhancing both interpretability and robustness. Experiments at 3&#xa0;ms show a range detection error below 0.05&#xa0;ms, with root mean square errors of 0.009&#xa0;Hz and 0.04&#xa0;Hz for RR and RR, respectively, over 60% more accurate than traditional fast Fourier transform methods. The approach is computationally efficient and highly applicable in telemedicine, smart home monitoring, and related fields.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A multi-target physiological signal detection method for UWB radar based on Kalman tracking and dual-branch network

  • 期刊论文

  • 发表日期：

    2025年11月1日

查看全文
  • Ziqi Li,
  • Dongyao Jia,
  • Zihao He,
  • Nengkai Wu

摘要

This paper introduces a novel non-contact physiological signal detection method using Ultra-Wideband (UWB) radar, integrating Kalman tracking and time-frequency deep learning to mitigate range gate jitter and noise in multi-target scenarios. A dynamic threshold adaptive peak detection method, combined with multi-Kalman filtering, ensures robust target acquisition. A deep learning model with frequency domain enhancement improves respiratory rate (RR) and heart rate (HR) detection accuracy. The framework employs multi-task learning, optimizing RR regression, HR classification, enhancing both interpretability and robustness. Experiments at 3 ms show a range detection error below 0.05 ms, with root mean square errors of 0.009 Hz and 0.04 Hz for RR and RR, respectively, over 60% more accurate than traditional fast Fourier transform methods. The approach is computationally efficient and highly applicable in telemedicine, smart home monitoring, and related fields.