Visual Multi-spectral Semantic Analysis and Prediction using Unmanned Vehicles

  • Ahmad S. Obeid

Student thesis: Master's Thesis

Abstract

The acquisition of data in the recent years has witnessed a great technological advancement. Various sensors are being automatically deployed, which collect heterogeneous data, spanning large portions of the light spectrum, to fine levels of resolution, as well as 3-dimensional (3D) information, at different periods of time. Such enrichment in the data query from the sensing side must be accompanied by a parallel and equivalent advancement in the analysis of the data. This is because as the data becomes more informative and comprehensive, so becomes the issue of handling and analyzing it more delicate and complex. For that, we find in machine learning-based methods a great tool for the different applications of remote sensing. However, the use of machine learning towards remote sensing comes with its own issues and considerations. In this thesis, we focus on two of these issues by studying them closely and proposing novel ways of approaching them. The first issue arises when the data-under-test is representatively skewed towards a category over the other(s), which is known as data-imbalance. The other issue is the automation of analyzing the remote sensing datasets, which continuously grows in importance as the acquisition of data evolves and the data becomes plenty. In total, four contributions are provided in this thesis, tackling the two mentioned focal points. A thorough review of the existing methods that tackle the two issue is conducted, and the proposed contributions are explained in detail and tested on different remote sensing datasets to prove their efficacy, where significant improvements were observed over the existing methods. Moreover, the applicability of the developed methods on another related machine learning task is demonstrated to showcase their generalizability.
Date of AwardMay 2020
Original languageAmerican English

Keywords

  • Remote sensing; Land cover classification; Change detection; Automatic and powerful machine learning paradigms; Challenging analysis complexities

Cite this

'