Expert fuzzy/holistic classifier applied to hyperspectral data over precious metal prospects in the SWIR (2.0 to 2.4 um)

A method is presented for the classification of hyperspectral imagery. The
method strives to deliver a product in which a typical user acquires concrete evidence
for believing the correctness of the classification. We discuss an example using the easiest
possible case so that the presentation is clear: mapping alunite over Cuprite,
Nevada, where mineralogical exposures are relatively unobscured and well defined.
The method brings together in a single package the two main methodologies that have
been used to classify images:
• the holistic approach in which each spectral wavelength is weighted equally, and
• a feature-based approach in which the identification relies on the presence of
recognizable spectral absorption features while ignoring the broad characteristics
of the image spectrum.
The increased availability of hyperspectral imagery using 200 or more narrow
spectral bands permits the increased use of accurate feature-based methods to distinguish
similar clay minerals in a way not possible from such imagery as thematic mapper
(TM). While nonexperts can interpret the results of this method, an expert is
required for the design of the classification except in the most straightforward cases.