Small UAS Online Audio DOA Estimation and Real-Time Identification Using Machine Learning.

Area of Science:

• Aerospace Engineering
• Acoustic Signal Processing
• Machine Learning Applications

Background:

• Increasing prevalence of unmanned aerial systems (UAS) necessitates effective detection and mitigation strategies.
• Existing counter-UAS (C-UAS) solutions often rely on complex sensor networks (radar, RF, EO/IR, acoustic).
• There is a need for accessible, cost-effective, and easily deployable UAS detection systems.

Purpose of the Study:

• To introduce a novel, small UAS (sUAS) acoustic detection system.
• To enable direction of arrival (DOA) estimation and acoustic signal identification for sUAS.
• To validate the system's performance in real-world outdoor conditions.

Main Methods:

• Utilizes a microphone array for continuous audio data collection.
• Implements a triangulation-like algorithm for direction of arrival (DOA) estimation.
• Employs machine learning (ML) techniques on sound spectrograms for sUAS identification.

Main Results:

• Successfully demonstrated direction of arrival (DOA) estimation of acoustic signals.
• Achieved reliable identification of acoustic signals originating from sUAS.
• Validated effective sUAS detection at distances exceeding 70 meters in outdoor experiments.

Conclusions:

• The developed acoustic system offers a viable, cost-effective solution for sUAS detection.
• The combination of DOA estimation and ML-based identification enhances detection reliability.
• The system's ease of deployment and performance make it suitable for various C-UAS applications.