Subpixel approaches to multispectral and hyperspectral image classification and application
Loading...
Links to Files
Permanent Link
Author/Creator
Author/Creator ORCID
Date
2019-01-01
Type of Work
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.
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
One of major advantages of hyperspectral image processing is subpixel detection that can be used to detect targets at subpixel scale (i.e., subtarget) which cannot be visualized by human eye inspection. This dissertations explores a potential application of subtarget detection in hyperspectral image classification (HSIC). It extends a well-known subpixel target detection algorithm, constrained energy minimization (CEM) to an iterative CEM (ICEM) using band selection and nonlinear expansion, called BSNE-ICEM. In order to capture spatial information for BSNE-ICEM, a Gaussian filter is included to be implemented in conjunction with CEM to generate Gaussian filtered CEM maps which are further fed back to be added to the current image cube to create a new image cube for next round processing of CEM. The same procedure is repeatedly in an iteration manner until it meets a custom-designed stopping rule. The BSNE-ICEM is then extended to its progressive version in hyperspectral image classification, called progressive hyperspectral image classification (PHSIC) which processes image classification in a progressive manner. Its idea converts a hyperspectral image classification problem to a series of hyperspectral detection problems via a hierarchical binary tree which can be constructed according to a new concept, called class classification difficulty (CCD). By virtue of this CCD-constructed binary tree only binary decisions are needed to be made at each node of a tree. When the progressive process reaches the last level (i.e., leave level), the classification is completed. To evaluate the hyperspectral image subpixel detection performance, the commonly used evaluation tool for target detection performance, receiver operating characteristic (ROC) analysis, is extended to 3-dimensional ROC (3D ROC) analysis for hyperspectral image classification to explore interrelationship among the detection probability (PD), false alarm probability (PF) and the threshold ?. As a potential application, BSNE-ICEM is applied to estimation of phosphorus concentration from inland water body using satellite images.