Automatic Life Detection Based on Efficient Features of Ground-Penetrating Rescue Radar Signals.

Area of Science:

• Engineering
• Signal Processing
• Biomedical Engineering

Background:

• Accurate classification in challenging scenarios like detecting breathing under rubble using bioradar is crucial for search and rescue.
• Bioradar operates under complex conditions, differing significantly from patient monitoring.
• Distinguishing between measurements with and without a person requires robust feature extraction.

Purpose of the Study:

• To identify optimal features for classifying bioradar signals indicative of human breathing under rubble.
• To find features robust to variations in rubble sites, breathing rates, and signal strength.
• To enhance the accuracy of life detection in disaster scenarios.

Main Methods:

• Decomposed bioradar signals into range intervals to create virtual scenes.
• Extracted physical and statistical quantitative features.
• Utilized Analysis of Variance (ANOVA) and Minimum Redundancy Maximum Relevance (MRMR) for feature significance analysis.
• Trained a linear kernel support vector machine (SVM) classification model.

Main Results:

• Identified an optimal feature set comprising four features.
• These features are related to distinct physical quantities.
• The selected features demonstrated independence from different rubble site conditions.
• Achieved improved classification accuracy using the optimal feature set.

Conclusions:

• An optimal set of four bioradar features has been identified for detecting breathing under rubble.
• These features are robust to environmental complexities encountered in rescue scenarios.
• The findings contribute to improved life detection capabilities in search and rescue operations.