A Deep Learning Method of Human Identification from Radar Signal for Daily Sleep Health Monitoring.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Radar signals offer a promising, non-contact method for human identification.
• Challenges exist in applying radar-based identification in uncontrolled home sleep-monitoring settings due to motion variability and data quality.
• Open-set recognition for radar sequences remains an under-explored area.

Purpose of the Study:

• To develop a robust deep learning model for human identification using radar sequences captured during sleep in daily home environments.
• To address the challenge of open-set recognition for radar-based human identification.
• To improve the accuracy and reliability of human identification systems in non-laboratory settings.

Main Methods:

• A deep convolution neural network (CNN) was employed for human identification from preprocessed radar sequences.
• Radar sequences were preprocessed to mitigate environmental interference, enhancing system robustness.
• A Principal Component Space feature representation was introduced for the detection of unknown (unseen) sequences.

Main Results:

• The proposed method achieved high labeling accuracies of 98.2% and 96.8% on average across public and experimental datasets, respectively.
• The system demonstrated superior performance compared to existing state-of-the-art techniques.
• Near 100% detection of unknown sequences was achieved with minimal misclassification of known individuals.

Conclusions:

• The deep learning approach effectively performs human identification from radar sequences in home sleep-monitoring scenarios.
• The method successfully addresses the open-set recognition problem, distinguishing known from unknown individuals.
• This technique offers a robust and accurate solution for non-contact human identification in real-world environments.