Self-Organizing Maps Applied to Mineral Spectra: Pure and Mixed

Details Self-Organizing Maps Applied to Mineral Spectra: Pure and Mixed Self-Organizing Maps Applied to Mineral Spectra: Pure and Mixed

Dec 2002

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2002agufm.b61a0711h&link_type=abstract

American Geophysical Union, Fall Meeting 2002, abstract #B61A-0711

Mathematics

Logic

0400 Biogeosciences, 5410 Composition, 5464 Remote Sensing, 5494 Instruments And Techniques

Scientific paper

Meaningful classification of spectral data is an important first step in mineralogical analysis of terrestrial and planetary surfaces. The large data sets being gathered (Tbytes) can not be manually evaluated in a timely, efficient manner. This suggests automated classification and interpretative processing must be developed and applied. Spectral libraries, databases of simulated or experimental spectra, play a crucial role in the identification of remotely obtained spectral data. At infrared wavelengths, linear mixing models are widely used to ascertain the principle mineral components contained in an unknown sample. An exhaustive search for spectral library candidates epresenting these components for a large database of samples is very CPU intensive, especially as the number of required components grows. A method to reduce the set of candidates in a linear mixing analysis would greatly improve the efficiency of the identification process. Classifications related to chemical and structural properties which are manifested in spectral features can be useful in this regard. We have developed a scheme based upon self-organizing maps (SOM) that organizes spectra in an unsupervised fashion into a 2-D map. This scheme was applied to the Arizona State University (ASU) thermal emission mineral spectral library and separated the samples into clusters that relate to distinct chemical and structural groups. Three hierarchical levels of distinction arise from the SOM classification; mineral class (e.g. silicates vs. oxides), mineral subclass (e.g. inosilicates vs. nesosilicates) and mineral group (e.g. pyroxenes vs. amphiboles). In addition, we investigated spectral mixing such as occurs in natural samples. An 'unknown' spectrum, constructed from a linear mix of ASU library spectra, is 'classified' by locating it in the 2-D SOM map. Through an iterative process the end members composing the unknown mixture were recovered by evaluating the members of the library that were closest in proximity to the unknown on the 2-D SOM map. We will show our SOM classifications of the ASU library and illustrate, with several mixture examples, how the SOM map is used to zero in on library end members.

Affiliated with

Also associated with

No associations

Rating

Self-Organizing Maps Applied to Mineral Spectra: Pure and Mixed does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Self-Organizing Maps Applied to Mineral Spectra: Pure and Mixed, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Self-Organizing Maps Applied to Mineral Spectra: Pure and Mixed will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-1891728