Share Email Print

Proceedings Paper

Segmenting clouds from space: a hybrid multispectral classification algorithm for satellite imagery
Format Member Price Non-Member Price
PDF $17.00 $21.00
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

This paper reports on a novel approach to atmospheric cloud segmentation from a space based multi-spectral pushbroom satellite system. The satellite collects 15 spectral bands ranging from visible, 0.45 um, to long wave infa-red (IR), 10.7um. The images are radiometrically calibrated and have ground sample distances (GSD) of 5 meters for visible to very near IR bands and a GSD of 20 meters for near IR to long wave IR. The algorithm consists of a hybrid-classification system in the sense that supervised and unsupervised networks are used in conjunction. For performance evaluation, a series of numerical comparisons to human derived cloud borders were performed. A set of 33 scenes were selected to represent various climate zones with different land cover from around the world. The algorithm consisted of the following. Band separation was performed to find the band combinations which form significant separation between cloud and background classes. The potential bands are fed into a K-Means clustering algorithm in order to identify areas in the image which have similar centroids. Each cluster is then compared to the cloud and background prototypes using the Jeffries-Matusita distance. A minimum distance is found and each unknown cluster is assigned to their appropriate prototype. A classification rate of 88% was found when using one short wave IR band and one mid-wave IR band. Past investigators have reported segmentation accuracies ranging from 67% to 80%, many of which require human intervention. A sensitivity of 75% and specificity of 90% were reported as well.

Paper Details

Date Published: 22 August 2005
PDF: 9 pages
Proc. SPIE 5882, Earth Observing Systems X, 58820I (22 August 2005); doi: 10.1117/12.617656
Show Author Affiliations
Mark Wilson, Sandia National Labs. (United States)
Prabal Nandy, Sandia National Labs. (United States)
Brian Post, Sandia National Labs. (United States)
Jody Smith, Sandia National Labs. (United States)
Joseph Wehlburg, Sandia National Labs. (United States)

Published in SPIE Proceedings Vol. 5882:
Earth Observing Systems X
James J. Butler, Editor(s)

© SPIE. Terms of Use
Back to Top