Deep Learning Classifier for Robust Artifact Rejection in FMCW Radar Vital Sensing

Accurate and continuous monitoring of vital signs like heart and breathing rate is essential for timely diagnosis and early detection of critical health conditions. While traditional wearable devices fulfill this purpose, they can be uncomfortable and impractical for continuous use. In this study, we propose a novel classifier for radar-based vital sensing that leverages deep learning to enhance the accuracy and reliability of noncontact vital sign detection. Our classifier, integrated as a body motion rejection block, effectively differentiates between high-quality and corrupted data. Experimental results demonstrate a significant performance improvement over conventional methods. Specifically, our approach achieves 99.1% accuracy, 98.9% precision, 99.5% recall, and a 99.2% <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${F}1$ </tex-math></inline-formula> score. These metrics represent an improvement of 8.4% in accuracy, 7.9% in precision, 6.7% in recall, and 7.3% in <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${F}1$ </tex-math></inline-formula> score compared to the reference algorithm, which records respective values of 90.7%, 91.0%, 92.8%, and 91.9%. Additionally, we explored several architectural variants, including the use of Range FFT, Doppler FFT, and their combination as inputs. Our findings indicate that using both Range FFT and Doppler FFT inputs yields optimal performance. The combination effectively captures both spatial and velocity information, leading to improved model accuracy and robustness. The enhanced performance of our classifier holds promise for its application in real-time, privacy-sensitive Internet of Things (IoT) environments, marking a significant advancement in the field of noncontact vital sign monitoring.