Swinney, Carolyn J. (University of Essex), Woods, John C. (University of Essex)

RF Detection and Classification of Unmanned Aerial Vehicles in Environments with Wireless Interference

Scheduled for presentation during the Regular Session "UAS Communications" (FrD2), Friday, June 18, 2021, 16:20−16:40, Kozani

2021 International Conference on Unmanned Aircraft Systems (ICUAS), June 15-18, 2021, Athens, Greece

This information is tentative and subject to change. Compiled on April 19, 2024

Abstract

Unmanned Aerial Vehicle (UAV) detection and classification methods include the use of audio, video, thermal, RADAR and radio frequency (RF) signals. RF signals have the ability to detect UAVs at longer ranges but interference from other signals in the same frequency band such as Bluetooth and Wi-Fi at 2.4GHz is a known limitation. The experiments in this paper evaluate the effect of real world Bluetooth and Wi-Fi signal interference on UAV detection and classification, using transfer learning via Convolutional Neural Network (CNN) feature extraction and machine learning classifiers Logistic Regression (LR) and k Nearest Neighbour (kNN). 2 class UAV detection, 4 class UAV type and 10 class flight mode classification are evaluated with graphical representation from the time and frequency domain. Flight modes evaluated included mode 1 - switched on and connected to the controller, mode 2 - hovering and mode 3 - flying. Results show that Bluetooth signals are more likely to interfere with detection and classification accuracy than Wi-Fi signals but that accuracy can be maintained at over 96% by using frequency domain features with LR as the classifier. Time domain features were shown to be less robust than frequency domain features when interference signals were introduced. In the presence of Bluetooth or Wi-Fi signals, 2 class UAV detection produced 100% accuracy, 4 class UAV type classification produced 99.9% (+/- 0.1%) and 10 class UAV flight mode classification produced 96.4% (+/- 0.5%) accuracy. Overall we have shown frequency domain features extracted from a CNN to be more robust than time domain features in the presence of interference and that high accuracy can be maintained using LR as a classifier with CNN derived features.