Millimeter-Wave Radar-Based Heart Rate Detection Method Using Adaptive Transformer Model During Exercise

Abstract

Heart rate (HR) detection, as an important physiological monitoring technique, has been widely adopted in clinical care and health management due to its non-invasive nature and continuity. Currently, most radar-based heart rate detection methods require the subject to remain stationary and process radar echoes via phase analysis, yet they fail to handle radar signals under exercise conditions. This study proposes a non-contact millimeter-wave radar heart rate detection method based on a power-aware adaptive self-attention transformer (ASAT). First, a motion power detection module is constructed to establish a motion–heart rate data correlation model. Then, an innovative integration of a channel-partitioned attention refinement feedforward network and Efficient Channel Attention (ECA) is implemented to effectively suppress motion artifacts and enhance critical features. Finally, heart rate is estimated based on power data. To validate the model's effectiveness, heart rate data under three different types of exercise were collected. The results show that the mean absolute error of heart rate estimation via power data is approximately 9.24 beat per minute (bpm), which is 4.05 bpm lower than that of Long Short-Term Memory (LSTM) models. This study establishes a novel paradigm for continuous health monitoring in complex exercise environments, offers an effective solution for radar technology applications in health monitoring, and lays a foundation for future research.