Physics-Aware Lightweight Deep Learning Model for Human Breathing Detection via UWB Radar
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Turkey, 7 - 10 July 2026, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/siu71813.2026.11636904
- City: İstanbul
- Country: Turkey
- Keywords: convolution neural network, human detection, UWB radar
- Yıldız Technical University Affiliated: Yes
Abstract
A physics-aware deep learning framework for human detection from raw ultra-wideband (UWB) radar signals is proposed in this paper. The proposed framework is based on a lightweight 2D Convolutional Neural Network (CNN) architecture that can be deployed on edge devices with limited computational resources. To better capture spatial-temporal and physiological patterns related to human respiration, physics-inspired features, including the Hilbert Envelope and Breathing Band Energy, are added as additional input channels to direct the network to concentrate on signal variations relevant to the presence of a human. In addition, model quantization methods are employed to drastically reduce the computational load and memory footprint, making it feasible to deploy on low-power embedded platforms. Experimental results show that the proposed framework achieves high detection accuracy with fast inference and minimal memory requirements.