Abstract: Addressing challenges such as low performance, high data signal-to-noise ratio requirements, and limited real-time capabilities in existing heart rate detection methods based on millimeter wave radar, this study presents a heart rate sensing approach tailored for weak vital sign signals characterized by low signal-to-noise ratio and missing data. The method applies a signal mask for echo sequences with variable length. Building upon this signal mask, a signal mapping technique that leverages morphology is devised to mitigate interference and noise. Additionally, learnable position encoding is incorporated to capture temporal features within the signal. Subsequently, a transformer encoder module is employed for matching and computation, culminating in the development of a time-series global regression model based on deep learning framework. Following the preparation of the dataset and model training, the proposed approach is validated by performance analysis experiments, interference resistance tests, and comparative experiments. Results indicate that this method achieves an impressive accuracy of 96.30% within signal durations ranging from 2 s to 5 s, and it is suitable for scenarios involving missing data and noise interference. Importantly, this approach effectively enables a precise heart rate sensing from short-duration radar signals.

Key words: millimeter wave radar, heart rate sensing, global regression model

摘要： 针对基于毫米波雷达的心率检测方法性能较低、数据信噪比要求较高、实时性不高等问题，设计了针对低信噪比和带有数据缺失的生命体征微弱信号的心率感知方法。首先对变长回波序列设计信号掩膜，在此基础上针对干扰噪声设计基于回波形态的信号映射方法，并添加可学习的位置编码以表征信号时序特征，然后通过transformer编码器模块匹配计算，最终构建了基于深度学习框架的时间序列全局回归模型。完成数据集制备及模型训练后，经过性能分析实验、抗干扰能力实验、对比实验验证，上述方法可在2~5 s的较短信号时间段内达到96.30%的准确率，同时适用于数据缺失及噪声干扰场景，能有效地实现短时雷达回波信号的心率体征参数精确感知。

关键词: 毫米波雷达，心率感知，全局回归模型