Please use this identifier to cite or link to this item:
https://www.um.edu.mt/library/oar/handle/123456789/120581
Title: | Enhancing UAV detection through machine learning utilising LTE radio measurements |
Authors: | Fenech, Kristian (2023) |
Keywords: | Drone aircraft -- Identification Machine learning MATLAB Support vector machines |
Issue Date: | 2023 |
Citation: | Fenech, K. (2023). Enhancing UAV detection through machine learning utilising LTE radio measurements (Master's dissertation). |
Abstract: | The main aim of the dissertation is to detect UAVs (Unmanned Aerial Vehicles) based upon the implementation of a machine learning method. In order to achieve this ultimate objective, a UAV simulator is implemented on MATLAB which provides a flexible environment whereby UAV flight paths can be simulated at a number of different heights. The simulator is employed to extract a variety of LTE radio measurements from shape files supplied by EPIC, a prominent telecommunications company in Malta. These measurements pertain to a drone simulation conducted at different altitudes, as well as at ground level. A number of features extracted from the shape files are then used to train three different machine learning algorithms which include Support Vector Machines, a Multi-Layer Perceptron model and the Random Forest algorithm. A number of different test cases are designed for different heights, as well as some cases that increase the amount of features used in order to ascertain whether there will be any effects on the results produced. These machine learning algorithms are then evaluated on the basis of a number of KPIs (Key Performance Indicators) to determine how good they are at detecting drones. Some of these KPIs include specificity and sensitivity. Using three features to train the machine learning models, the Multi-Layer Perceptron model produces the most accurate results, with a sensitivity score of up to 95% in certain scenarios. When the number of features increases to five from three, a number of improvements are noted for the Random Forest and Support Vector Machine algorithms, however the Multi-Layer Perceptron model is shown to regress, which warrants further investigation. |
Description: | M.Sc. (Melit.) |
URI: | https://www.um.edu.mt/library/oar/handle/123456789/120581 |
Appears in Collections: | Dissertations - FacICT - 2023 Dissertations - FacICTCCE - 2023 |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
.PDF Restricted Access | 3.76 MB | Adobe PDF | View/Open Request a copy |
Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.