Date of Award

2018-01-01

Degree Name

Doctor of Philosophy

Department

Electrical Engineering

Advisor(s)

Miguel Velez-Reyes

Abstract

Unsupervised unmixing analysis aims to extract the basic materials, also known as, endmembers, and their fractions (or abundances) from a hyperspectral image. In this work, a new unsupervised unmixing approach based on Low Dimensional representation with regional analysis is developed. A low dimensional image representation is obtained using superpixel segmentation where each superpixel is represented by its mean spectra. Regional analysis is then performed using the quadtree segmentation based on the Shannon entropy. Endmembers are extracted from each region and combined using clustering into endmember classes. The proposed approach is tested and validated using the HyDICE Urban, ROSIS Pavia, and AVIRIS Fort AP Hill data sets. Different levels of qualitative and quantitative assessments are performed based on the available reference data. The proposed approach is also compared with global (no-regional segmentation) and with pixel-based (no-superpixel dimensionality reduction) unsupervised unmixing approaches. Qualitative assessment was based primarily on agreement with spatial distribution of materials provided by a reference classification map. Quantitative assessment was based on comparing classification maps derived from abundance maps using winner takes it all and reference data. High agreements with reference data were obtained by the proposed approach as evidenced by high kappa values (over 80%). The proposed approach outperforms all global unsupervised unmixing approaches that do not account for regional information. Performance is slightly better when compared to pixel-based approached using regional segmentation, however our approach resulted in significant computational savings due to the dimensionality reduction step.

Language

en

Provenance

Received from ProQuest

File Size

128 pages

File Format

application/pdf

Rights Holder

Mohammed Qassim Alkhatib

Download

Share

COinS