A Compressive Sensing Approach To Hyperspectral Image Classification

Author/Creator ORCID

Date

2019-01-01

Department

Computer Science and Electrical Engineering

Program

Engineering, Electrical

Citation of Original Publication

Rights

Distribution Rights granted to UMBC by the author.
Access limited to the UMBC community. Item may possibly be obtained via Interlibrary Loan thorugh a local library, pending author/copyright holder's permission.
This item is likely protected under Title 17 of the U.S. Copyright Law. Unless on a Creative Commons license, for uses protected by Copyright Law, contact the copyright holder or the author.

Abstract

Hyperspectral imaging (HSI) technology has found success in a variety of applications; however, its use is often still limited due to size, weight and power (SWaP) constraints. In this dissertations, compressive sensing (CS) is proposed as an enabling technology to reduce the high spectral band count, through the creation of compressively-sensed bands (CSBs). A CS model based on the universality of random sensing is proposed for the analysis of hyperspectral classification in the compressed domain. Specifically, the utility of the support vector machine (SVM) in the compressed domain is evaluated through both mathematical analysis and empirical experimentation. This work shows that is indeed possible to achieve full band classification performance in the compressed domain. The experiments also demonstrate that the minimum number of CSBs is scene dependent, requiring additional algorithms to provide a full solution. Two supervised algorithms based on a feature selection framework are proposed for estimating the minimum lower bound on the required number of CSBs. The first algorithm is based on feature filtering techniques and the second algorithm is based on classifier wrapping. Finally, an unsupervised algorithm is presented, based on progressive band processing, that is able to adaptively determine the required number of CSBs in-situ. The contributions of this dissertations provide the fundamental foundation for compressed classification of hyperspectral images and identify several new opportunities for future research.