Date on Master's Thesis/Doctoral Dissertation
Computer Engineering and Computer Science
Computer Science and Engineering, PhD
Committee Co-Chair (if applicable)
Wavelet scattering transform; continuous wavelet transform; wavelet packet transform; local outlier factor; denoising autoencoder
Advancement in technology has led to creative and innovative inventions. One such invention includes unmanned aerial vehicles (UAVs). UAVs (also known as drones) are now an intrinsic part of our society because their application is becoming ubiquitous in every industry ranging from transportation and logistics to environmental monitoring among others. With the numerous benign applications of UAVs, their emergence has added a new dimension to privacy and security issues. There are little or no strict regulations on the people that can purchase or own a UAV. For this reason, nefarious actors can take advantage of these aircraft to intrude into restricted or private areas. A UAV detection and identification system is one of the ways of detecting and identifying the presence of a UAV in an area. UAV detection and identification systems employ different sensing techniques such as radio frequency (RF) signals, video, sounds, and thermal imaging for detecting an intruding UAV. Because of the passive nature (stealth) of RF sensing techniques, the ability to exploit RF sensing for identification of UAV flight mode (i.e., flying, hovering, videoing, etc.), and the capability to detect a UAV at beyond visual line-of-sight (BVLOS) or marginal line-of-sight makes RF sensing techniques promising for UAV detection and identification. More so, there is constant communication between a UAV and its ground station (i.e., flight controller). The RF signals emitting from a UAV or UAV flight controller can be exploited for UAV detection and identification. Hence, in this work, an RF-based UAV detection and identification system is proposed and investigated. In RF signal fingerprinting research, the transient and steady state of the RF signals can be used to extract a unique signature. The first part of this work is to use two different wavelet analytic transforms (i.e., continuous wavelet transform and wavelet scattering transform) to investigate and analyze the characteristics or impacts of using either state for UAV detection and identification. Coefficient-based and image-based signatures are proposed for each of the wavelet analysis transforms to detect and identify a UAV. One of the challenges of using RF sensing is that a UAV's communication links operate at the industrial, scientific, and medical (ISM) band. Several devices such as Bluetooth and WiFi operate at the ISM band as well, so discriminating UAVs from other ISM devices is not a trivial task. A semi-supervised anomaly detection approach is explored and proposed in this research to differentiate UAVs from Bluetooth and WiFi devices. Both time-frequency analytical approaches and unsupervised deep neural network techniques (i.e., denoising autoencoder) are used differently for feature extraction. Finally, a hierarchical classification framework for UAV identification is proposed for the identification of the type of unmanned aerial system signal (UAV or UAV controller signal), the UAV model, and the operational mode of the UAV. This is a shift from a flat classification approach. The hierarchical learning approach provides a level-by-level classification that can be useful for identifying an intruding UAV. The proposed frameworks described here can be extended to the detection of rogue RF devices in an environment.
Medaiyese, Olusiji Oloruntobi, "Signal fingerprinting and machine learning framework for UAV detection and identification." (2021). Electronic Theses and Dissertations. Paper 3687.