Finding Endmember Classes and Endmembers in Hyperspectral Images
Loading...
Links to Files
Permanent Link
Collections
Author/Creator
Author/Creator ORCID
Date
2016-01-01
Type of Work
Department
Computer Science and Electrical Engineering
Program
Engineering, Electrical
Citation of Original Publication
Rights
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
Distribution Rights granted to UMBC by the author.
Distribution Rights granted to UMBC by the author.
Abstract
Remotely sensed images have been used in a broad range of applications ranging from chemical/biological defense, geology, agriculture to environmental protection, law enforcement and intelligence applications. With the recent advanced technology, remote sensing instruments have significantly improved spatial resolution and also spectral resolution. The images from these instruments have hundreds of different contiguous spectral bands with typical spectral resolution of approximately 10 nanometers which can be used to uncover subtle material substances that cannot be resolved by multispectral sensors. With significantly improved spectral and spatial resolutions provided by hyperspectral imaging sensors, one challenging issue is to find how many spectrally distinct signatures are present in the resulting image data. Recently, Virtual Dimensionality (VD) was developed to address this issue which is defined as the number of spectrally distinct signatures in hyperspectral imagery. Once VD is determined, a follow-up issue is to find these signatures from the image data. As defined, an endmember is an ideal, pure signature for a class, more specifically, spectral class. However, in reality, the spectral signature for a material may vary due to a number of reasons including environmental, atmospheric and temporal factors. As a result, an endmember may appear in various forms in hyperspectral images. In order to resolve endmember variability issue, this dissertations develops three new algorithms. The first one is called Endmember Variability Algorithm (EVA) which is an unsupervised endmember class finding algorithm using a half-way A second one develops a new criterion similar to Fisher's ratio used in Fisher's Linear Discriminant Analysis (FLDA) and various versions of algorithms using the new criterion are further designed. Unlike current methods that either require training data set or need a predefined parameter, two proposed methods can find endmembers and their corresponding classes in a complete unsupervised manner. A third algorithm is Fully Constrained Least Square Endmember Finding Algorithm (FCLS EFA) which uses average unmixing error to find endmembers that assumed to be most representative pixels as endmembers for linear spectral representation via a linear mixing model.