Classification of RF Fingerprint Signals from UAV Controller Using Machine Learning Techniques

Unmanned aerial vehicles (UAVs) have gained widespread popularity in various industries with advent in technology due to their advantages and cost-effectiveness, but they also pose serious threats to security if misused. Detection and classification of radio frequency (RF) signals obtained through communication between a drone and its controller is one of the ways to look out for and prevent misapplication of UAVs. This paper presents a novel approach to classify RF signals from distinct drones using machine learning (ML) techniques. The study involves training and evaluating of multiple ML models on a limited batch of dataset consisting of 5 specimens of RF signals from various drone types. In addition, the paper also identifies the key features of signals that are needed to be extracted from RF signal data for appropriate modeling and classification. The key objective of this research was to achieve best accuracy while using limited dataset by identifying and including most valuable features of RF signal. Important signal features like Mean, Variance, Skewness, Kurtosis and Energy Spectral Entropy were extracted from the RF data. Despite the minimal data used, the kNN classification achieves the accuracy of 88.2% with inclusion of all 17 classes of drones and improved to 92.86% when trained on 14 classes. This shows promising potential for real-world application. The study demonstrates that even with a small training dataset, ML techniques can effectively classify RF signals from drones.