Malicious UAV Detection Using Blind Source Separation Algorithm and Neural Network Classifier

Abstract

Unmanned aerial vehicle(UAV) technology is the rapid growing technology in the field of monitoring for security purposes, pesticides spraying and various other applications. In the recent days, one of the major concerns is entering of malicious UAVs into the secured perimeter that might result in Drone-based cyberattacks. So, the detection of these malicious UAVs are crucial. In this work, an acoustic method of detecting malicious UAVs is proposed. The mixed form of the acoustic signals of two kinds of drones, namely, Fixed- wing and Multi- rotor are passed through the Blind Source Separation (BSS) block, where the kurtosis is measured along with Independent Component Analysis (ICA) for the separation of the signals. Then the distinctive features, Mel-Frequency Cepstral Coefficient(MFCC), Gamma tone-Frequency Cepstral Coefficient(GTCC) and short time energy are extracted from the acoustic signal and are trained using Neural Network(NN) classifier to identify the malicious UAV. The proposed method under different conditions outperforms the existing techniques with an accuracy of 100% in identification of malicious UAV. Key Words: Blind Source Separation Algorithm, kurtosis, Independent Component Analysis, Mel-Frequency Cepstral Coefficient(MFCC), Gamma tone-Frequency Cepstral Coefficient(GTCC), short time energy, Neural Networks