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

Abstract

The wide range of unmanned aerial system (UAS) applications has led to a substantial increase in their numbers, giving rise to a whole new area of systems aiming at detecting and/or mitigating their potentially unauthorized activities. The majority of these proposed solutions for countering the aforementioned actions (C-UAS) include radar/RF/EO/IR/acoustic sensors, usually working in coordination. This work introduces a small UAS (sUAS) acoustic detection system based on an array of microphones, easily deployable and with moderate cost. It continuously collects audio data and enables (a) the direction of arrival (DOA) estimation of the most prominent incoming acoustic signal by implementing a straightforward algorithmic process similar to triangulation and (b) identification, i.e., confirmation that the incoming acoustic signal actually emanates from a UAS, by exploiting sound spectrograms using machine-learning (ML) techniques. Extensive outdoor experimental sessions have validated this system's efficacy for reliable UAS detection at distances exceeding 70 m.