Key points are not available for this paper at this time.
Small Unmanned Aerial Systems (UAS) usage is undoubtedly increasing at a significant rate. However, alongside this expansion is a growing concern that dependable low-cost counter measures do not exist. To mitigate a threat in a restricted airspace, it must first be known that a threat is present. With airport disruption from malicious UASs occurring regularly, low-cost methods for early warning are essential. This paper considers a low-cost early warning system for UAS detection and classification consisting of a BladeRF software-defined radio (SDR), wideband antenna and a Raspberry Pi 4 producing an edge node with a cost of under USD 540. The experiments showed that the Raspberry Pi using TensorFlow is capable of running a CNN feature extractor and machine learning classifier as part of an early warning system for UASs. Inference times ranged from 15 to 28 s for two-class UAS detection and 18 to 28 s for UAS type classification, suggesting that for systems that require timely results the Raspberry Pi would be better suited to act as a repeater of the raw SDR data, enabling the processing to be carried out on a higher powered central control unit. However, an early warning system would likely fuse multiple sensors. These experiments showed the RF machine learning classifier capable of running on a low-cost Raspberry Pi 4, which produced overall accuracy for a two-class detection system at 100% and 90.9% for UAS type classification on the UASs tested. The contribution of this research is a starting point for the consideration of low-cost early warning systems for UAS classification using machine learning, an SDR and Raspberry Pi.
Swinney et al. (Tue,) studied this question.