Unmanned Aircraft Vehicle (UAV) Pilot Identification Using Machine Learning

Unmanned Aircraft Vehicle (UAV) Pilot Identification Using Machine Learning
Ahmed Saeed Al Shemeili
PhD. Thesis
February 2019
A thesis submitted to Khalifa University of Science and Technology in accordance with the
requirements of the degree of PhD in Engineering in the Department of (Electrical and
Computer Engineering).
1

Abstract
Ahmed Saeed Al Shemeili, “Unmanned Aircraft Vehicle (UAV) Pilot Identification Using
Machine Learning”, PhD Thesis, Program Name, Department of (Electrical and Computer
Engineering), Khalifa University of Science and Technology, United Arab Emirates,
February 2019.
Abstract—the objective is to capture the behavioral data of different pilots while flying the
UAV. Based on that data classification, the algorithm should be able to identify the pilot of
each dataset. 20 datasets were provided, each containing at least 2600 instance (example:
pitch, yaw, thrust, roll, time, pilot, etc.) at different flight modes (triangle, vertical, random).
Trained model will be generated that can identify individual pilot by feeding drone data. This
mechanism will enable the UAV to recognize the pilot in order to avoid being hijacked by
non-authorized pilot.
Drone flight time is not long due to power constrains and this needed to be taken into our
consideration during the design of an online pilot identification model using machine
learning. Machine learning algorithms needs time and power to process and classify datasets.
20 features were tested and classified against 22 classifiers, were Random Forest was selected
as the best classifier. Part of the 20 features comes from Radio Controlled (RC) signal and
from the Internal Measurement Unit (IMU) sensor. Primary results proofed that a single
feature can score an accuracy of 95.9 % in pilot identifications, were the overall performance
of combined features was 99.40%. This new finding will help in enhancing training
/detection time and processing power and memory of our model.
Random forest tree reduction will be evaluated against the generated model; were RF tree
will be examined on how it selects specific pilot for each tree. Objective will be to discover
specific pattern on the distribution of specific trees related to specific pilot and based on this
2

pattern, we will propose a reduction selection tree route that saves the power and memory for
not being running the data across the whole generated trees (only specific route).
Indexing Terms: UAV, machine learning, random forest, ensemble, bagging, MATLAB
3

Acknowledgement
Undertaking this PhD has been a major work experience change for me where I shift my
career from networking in cyber security and artificial intelligence and this was possible with
the support I received from many people and Khalifa University.
I would like first to thank my supervisor Dr. Abdulhadi Shoufan for all the support he gave
me and the continuous feedback, which helps during my PhD study.
Many thanks to Prof. Ernest Damiani for his support and guidance to ensure that my PhD
study will be successful and of benefit to our society
I gratefully acknowledge the funding received towards my PhD from Khalifa university.
Thanks to Prof. Mahmoud Al Qutairi for supporting my study at Khalifa University
Thanks to my Father, Mom, Brothers and sisters
Thank you, my Wife and kids,
4

Declaration and Copyright
Declaration
I declare that the work in this thesis was carried out in accordance with the regulations of
Khalifa University of Science and Technology. The work is entirely my own except where
indicated by special reference in the text. Any views expressed in the thesis are those of the
author and in no way represent those of Khalifa University of Science and Technology. No
part of the thesis has been presented to any other university for any degree.
Author Name: Ahmed Saeed Al Shemeili
Author Signature:
Date: 24-02-2019
5

Copyright ©
No part of this thesis may be reproduced, stored in a retrieval system, or transmitted, in any
form or by any means, electronic, mechanical, photocopying, recording, scanning or
otherwise, without prior written permission of the author. The thesis may be made available
for consultation in Khalifa University of Science and Technology Library and for inter-
library lending for use in another library and may be copied in full or in part for any bona fide
library or research worker, on the understanding that users are made aware of their
obligations under copyright, i.e. that no quotation and no information derived from it may be
published without the author's prior consent.
6

Contents
Abstract iii
Acknowledgement ii
Declaration and Copyright i
CHAPTER ONE 8
1.0 INTRODUCTION 8
1.2 TYPES OF MULTI-ROTORS 10
1.3 APPLICATIONS OF DRONES 11
Drones application in agriculture 11
Application of drones on the delivery of goods 14
Drones in fisheries, forestry and protection of wildlife 16
Drones in defense 17
Drones in Civil applications 17
Drones application in search and rescue 18
Application of drones in remote sensing 19
Drones application in the inspection of construction and infrastructure 20
Application of drones to monitor the traffic of roads 21
Application of drones to provide wireless coverage 22
Application of drones in the management of karst resources features 22
1.4 SECURITY ISSUES AND CHALLENGES POSED BY DRONES 23
7

1.5 DRONE BIOMETRICS AND PILOT IDENTIFICATION 26
1.6 MACHINE LEARNING TECHNIQUES 27
1.7 AIMS AND OBJECTIVE 29
1.9 CONCLUSION 31
CHAPTER TWO 33
2.1 WORKING OF UAVS 33
2.2 SECURITY ASPECTS IN UAVS 36
2.3 SAFETY ASPECTS - CYBER SECURITY APPROACHES 41
2.4 THUMBPRINT AND BIOMETRIC SECURITY DEVICES 51
2.5 MACHINE LEARNING APPROACHES 59
2.5 CONCLUSION 99
CHAPTER THREE 101
3.1 MACHINE LEARNING 101
3.2 MACHINE LEARNING TECHNIQUES 101
Supervised learning 102
Unsupervised Learning 102
3.3 CLASS IMBALANCE IN MACHINE LEARNING 103
Detect a “naive behavior” 103
Confusion matrix, precision, recall and F1 103
3.4 CRITICAL ANALYSIS 105
3.4.1 Several classifiers for further analysis 106
8