Hyperspectral Data

[Chris C Rewerts]

>224 bands of hyperspectral data:

This brings to mind some work that was done by Haluk Cetin, Tim Warner 
and Don Levandowski at Purdue Dept of Earth and Atmospheric Science.

If you are looking into classification of hundreds of bands, then 
you may want to check into their work. It has not been ported to
GRASS, but I believe it has been programmed in C on an IBM-PC version
of ERDAS.

The following blurb was an announcement to a seminar presentation
they made, and I am including it to serve as a brief description of
their work:

--
Topic:  nPDF Image Processing and Classification: AVIRIS and TIMS Applications

The nPDF analysis procedure is a powerful technique for the display,
enhancement and classification of high dimensional data, such as AVIRIS or
TIMS.  The technique is both user-interactive and extremely fast, making
it very effective for PC-based image processing.  An entire AVIRIS scene,
covered by 512 lines by 614 pixels and 180 bands, representing over 56
megabytes of data, was classified in under 13 minutes on a personal
computer.  On a smaller test area, using 15 AVIRIS bands, the nPDF
approach had a 71% classification accuracy, compared with 53% for minimum
distance, 67% for maximum likelihood, and 68% for Mahalanobis distance.

TIMS applications are illustrated through both image transformations and
classifications.  The standard method of TIMS display is a decorrelation
stretch, which tends to enhance noise, or Principal Components, which
is scene dependent and difficult to interpret.  nPDF transformation
images are scene independent, rapidly computed, do not enhance
noise and are relatively simple to  interpret.  The nPDF classification
technique is illustrated by the classification of TIMS six band emittance
data of Death Valley.

---

For further reference:

Cetin, Haluk and Donald Levandowski. Interactive Classification and 
Mapping of Multi-Dimensional Remotely Sensed Data Using n-Dimensional
Probability Density Functions (nPDF). Photogrammetric Engineering & 
Remote Sensing, Vol. 57, No. 12, December 1991, pp. 1579-1587.