Detection and classification of drones using Fourier-Bessel series representation of acoustic emissions

Abstract

Detection, classification, and line-of-sight range estimation of drones are vital for security, safety, and privacy reasons. Representation of the audio emissions of drones in a Fourier-Bessel (FB) series expansion is proposed for the identification of a drone and/or the prediction of its range from an observation point. A deep learning network employing the FB series coefficients as the preprocessed input has been shown to classify accurately each of seven drones flying in a controlled environment in about 84 % of cases. For the case of any one of three drones flying outdoors, presence of the drone—as opposed to background noise—was detected correctly with few false positive and false negative results. Additionally, the range of the drone—from 2.5 m to 935.6 m—was estimated to be within ±50 cm of actual line-of-sight distance in over 85 % of the available test cases.