COMPREHENSIVE DATABASE OF UAV SOUNDS FOR MACHINE LEARNING

Abstract

The use of Unmanned Aerial Vehicles (UAV) is steadily increasing. Besides the resulting benefits, there are also risks and dangers such as airspace violations or terrorist attacks, which require the development of effective drone defence systems. The realization of a drone defence system implies the following stages: Detection, Identification, Localisation and Neutralisation. In this paper, we address the drone detection and identification (classification) stage via acoustics using machine learning algorithms. A major problem with this approach is the lack of publicly available drone audio data. For this reason, we are building an extensive, open-access database consisting of both existing drone sounds and own drone recordings. This database contains drone sounds for all open drone classes from C0 (< 250 g) to C4 (< 25 kg).