Unsupervised unmixing of hyperspectral images : imaging Martian surface
Abstract/Contents
- Abstract
- Central scientific questions about Mars are whether any past environments were habitable, when and where they existed, and which geologic deposits could have preserved evidence of past life that developed in these environments. On Earth, the preservation of a record of life generally requires that life-forming materials be trapped fine-grained sediments, especially clays. The orbital observations of CRISM (Compact Reconnaissance Imaging Spectrometer for Mars) on the Mars Reconnaissance Orbiter (MRO) have dramatically improved the spatial resolutions of mineralogic evidence for past aqueous environments. CRISM collects electromagnetic radiation in the spectral range of 0.4-3.9 [mu]m at a resolution of 19 m/pixel. Each CRISM pixel is a 'spectrum' of radiance values as a function of wavelength. CRISM spectral resolution allowed the identification of several mineral species from their diagnostic spectral signature (the functional shape of their spectrum). Olivine and pyroxene (dominant phases in Mars' basaltic crust), phyllosilicates, hydrated silica, sulfates, oxyhydroxides, and carbonates have been identified. These identifications refer to the dominant mineral species present in a certain hyperspectral scene, because remotely observed spectra are most likely the result of spectral mixing. Spectral unmixing is the procedure by which the measured spectrum of a pixel is decomposed into a collection of constituent mineral spectra, or endmembers, and a set of corresponding fractions, or abundances, that indicate the proportion of each endmember present in the pixel. Unsupervised learning algorithms for unmixing lack the necessary validation in planetary data analysis due to the incomplete knowledge of the atmospheric and geologic variables. One of the goals of this work is to develop a complete simulation of the image formation process in Mars hyperspectral images in order to provide validation and testing for unmixing algorithms proposed in the thesis and in the literature for Mars scenarios. Spectral unmixing algorithms applied to CRISM datasets face particular challenges due to the peculiar behavior of the sensor noise, the presence of instrumental artifacts and the nonlinearity of the mixing. The main objective of this research is to develop a method to solve the unsupervised unmixing problem for CRISM hyperspectral images. The unmixing algorithm consists of different steps. Dimensionality reduction enhances the geometric distances between natural clusters in the data, facilitating the identification of endmembers. The techniques tends to produce a lower dimensional representation of the data with well separated components that are successively identified as natural clusters by a spectral clustering technique. The regions in the original space corresponding to the clusters are further analyzed by an unmixing algorithm that describe the roughly convex shapes of the regions with convex polyhedra whose vertices are local endmembers. Screening is then necessary in order to remove redundant spectra and obtain unique spectral signatures.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Publication date | 2010 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Parente, Mario |
|---|---|
| Associated with | Stanford University, Department of Electrical Engineering |
| Primary advisor | Gray, Robert M, 1943- |
| Thesis advisor | Gray, Robert M, 1943- |
| Thesis advisor | Bishop, Janice L |
| Thesis advisor | Tibshirani, Robert |
| Advisor | Bishop, Janice L |
| Advisor | Tibshirani, Robert |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Mario Parente. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis (Ph. D.)--Stanford University, 2010. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2010 by Mario Parente
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...