Spatiotemporal Feature Learning for Daily-Life Cough Detection Using FMCW Radar

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Cough detection is vital for respiratory health monitoring and disease management.
• Existing systems lack robustness due to limitations in modeling spatial and temporal cough data.
• Need for objective, reliable, and non-invasive cough detection methods.

Purpose of the Study:

• To develop a robust cough recognition framework using frequency-modulated continuous-wave (FMCW) radar.
• To integrate deep learning models for enhanced spatial and temporal feature extraction.
• To evaluate the system's performance in diverse, real-world conditions.

Main Methods:

• Utilized a frequency-modulated continuous-wave (FMCW) radar system for data acquisition.
• Developed a deep learning model combining a convolutional neural network (CNN) for spatial features and a Self-Attention mechanism for temporal dependencies.
• Employed data augmentation techniques to improve model generalization across various positions and activities.

Main Results:

• Achieved a high mean F1-score of 0.974±0.016 and accuracy of 99.05±0.55% under subject-independent cross-validation.
• Demonstrated strong performance with precision (98.77±1.05%), recall (96.07±2.16%), and specificity (99.73±0.23%).
• Validated the model's robustness in realistic scenarios with a large-scale dataset.

Conclusions:

• The proposed FMCW radar-based cough recognition framework is highly accurate and robust.
• This technology offers a practical solution for continuous, non-invasive, and privacy-preserving respiratory health monitoring.
• The system has significant potential for both clinical applications and telehealth services.