Entropy based Detection approach for Micro-UAV and Classification using Machine Learning

In this paper, we explore the techniques for detection and classification of Unmanned Aerial Vehicles (UAVs) using statistical features of the remote controller Radio Frequency (RF) signals in the presence of environmental noise. In the detection mechanism, the RF signal is transformed into Wavelet domain to filter out noise as well as to reduce computational cost. A kernel entropy based approach is used to partition the RF signal into bins and detect the presence of UAV. Unlike Conventional approaches, we compute the energy transient of signal from the Short Time Fourier Transform (STFT) coefficients obtained from Spectrogram of RF signal. Further, the higher order statistical features of energy transient signal are derived and ranked using Neighborhood Component Analysis (NCA)to select notable features for reducing the computational overhead. Finally, the significant features are used to train machine learning algorithm for classification. The algorithms are trained and tested using MPACT DroneRC Dataset containing 50 RF signals from each of the 15 different micro-UAV controllers. The dataset is partitioned with train to test ratio of 4:1 i.e., 80% of dataset is used for training and 20% for testing the algorithm. The k- Nearest Neighbor (kNN) algorithm with NCA classifies all micro-UAVs with an accuracy of 96.66%. The detection technique is also simulated for different Signal to Noise Ratio (SNR) levels and outcomes are reported.