Share Email Print

Proceedings Paper

Applying reconfigurable hardware to the analysis of multispectral and hyperspectral imagery
Author(s): Miriam E. Leeser; Pavle Belanovic; Michael Estlick; Maya Gokhale; John J. Szymanski; James P. Theiler
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Unsupervised clustering is a powerful technique for processing multispectral and hyperspectral images. Last year, we reported on an implementation of k-means clustering for multispectral images. Our implementation in reconfigurable hardware processed 10 channel multispectral images two orders of magnitude faster than a software implementation of the same algorithm. The advantage of using reconfigurable hardware to accelerate k-means clustering is clear; the disadvantage is the hardware implementation worked for one specific dataset. It is a non-trivial task to change this implementation to handle a dataset with different number of spectral channels, bits per spectral channel, or number of pixels; or to change the number of clusters. These changes required knowledge of the hardware design process and could take several days of a designer's time. Since multispectral data sets come in many shapes and sizes, being able to easily change the k-means implementation for these different data sets is important. For this reason, we have developed a parameterized implementation of the k-means algorithm. Our design is parameterized by the number of pixels in an image, the number of channels per pixel, and the number of bits per channel as well as the number of clusters. These parameters can easily be changed in a few minutes by someone not familiar with the design process. The resulting implementation is very close in performance to the original hardware implementation. It has the added advantage that the parameterized design compiles approximately three times faster than the original.

Paper Details

Date Published: 17 January 2002
PDF: 8 pages
Proc. SPIE 4480, Imaging Spectrometry VII, (17 January 2002); doi: 10.1117/12.453329
Show Author Affiliations
Miriam E. Leeser, Northeastern Univ. (United States)
Pavle Belanovic, Northeastern Univ. (United States)
Michael Estlick, Northeastern Univ. (United States)
Maya Gokhale, Los Alamos National Lab. (United States)
John J. Szymanski, Los Alamos National Lab. (United States)
James P. Theiler, Los Alamos National Lab. (United States)

Published in SPIE Proceedings Vol. 4480:
Imaging Spectrometry VII
Michael R. Descour; Sylvia S. Shen, Editor(s)

© SPIE. Terms of Use
Back to Top