Recover from Horcrux: A Spectrogram Augmentation Method for Cardiac Feature Monitoring from Radar Signal Components.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Contactless vital sign monitoring using radar is promising but hindered by data scarcity for deep learning.
• Existing data augmentation methods are often limited to classification tasks, not regression.

Purpose of the Study:

• To propose Horcrux, a spectrogram augmentation method for radar-based cardiac monitoring.
• To enable data augmentation for both classification and regression tasks in this domain.
• To improve deep learning model performance on limited datasets for vital sign monitoring.

Main Methods:

• Developed Horcrux, a spectrogram augmentation technique for radar signals.
• Ensured augmented spectrograms remain faithful to original vital sign data.
• Incorporated zero-value injection to highlight subtle cardiac features.

Main Results:

• Horcrux achieved a 16.20% overall improvement in cardiac monitoring.
• Demonstrated enhanced deep learning model performance with limited data.
• Showcased potential for application in other spectrogram-based tasks.

Conclusions:

• Horcrux effectively addresses data scarcity in deep learning for radar-based cardiac feature extraction.
• The method facilitates future clinical contactless vital sign monitoring with limited datasets.
• Horcrux offers a promising approach for enhancing deep learning in biomedical signal processing.