Robust Drone Classification Using Two-Stage Decision Trees and Results from SESAR SAFIR Trials

Abstract

Non-cooperative surveillance of drones is an important consideration in the EU SESAR vision for the provision of U-space services. The Aveillant Gamekeeper multiple beam staring radar utilises extended dwells to be able to detect small drones at ranges of several kilometres. However, target discrimination is necessary with such non-cooperative surveillance system as the increased detection sensitivity against low RCS targets, such as birds and surface objects (e.g., pedestrians and vehicles), extenuates the problem of false target reports. A simple two-stage supervised learning approach is proposed in order to discriminate drones from other confuser targets. This approach is based on a decision tree classifier and is shown to be effective at filtering out non-drone, targets. Field trials from the SESAR SAFIR trials to test initial U-space services in realistic urban environments shows that the two-stage decision tree classifier provides robust discrimination with minimal false positives.