Share Email Print
cover

Journal of Applied Remote Sensing

Evolving spectral transformations for multitemporal information extraction using evolutionary computation
Author(s): Henrique G. Momm; Greg Easson
Format Member Price Non-Member Price
PDF $20.00 $25.00
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

Remote sensing plays an important role in assessing temporal changes in land features. The challenge often resides in the conversion of large quantities of raw data into actionable information in a timely and cost-effective fashion. To address this issue, research was undertaken to develop an innovative methodology integrating biologically-inspired algorithms with standard image classification algorithms to improve information extraction from multitemporal imagery. Genetic programming was used as the optimization engine to evolve feature-specific candidate solutions in the form of nonlinear mathematical expressions of the image spectral channels (spectral indices). The temporal generalization capability of the proposed system was evaluated by addressing the task of building rooftop identification from a set of images acquired at different dates in a cross-validation approach. The proposed system generates robust solutions (kappa values > 0.75 for stage 1 and > 0.4 for stage 2) despite the statistical differences between the scenes caused by land use and land cover changes coupled with variable environmental conditions, and the lack of radiometric calibration between images. Based on our results, the use of nonlinear spectral indices enhanced the spectral differences between features improving the clustering capability of standard classifiers and providing an alternative solution for multitemporal information extraction.

Paper Details

Date Published: 1 January 2011
PDF: 19 pages
J. Appl. Remote Sens. 5(1) 053564 doi: 10.1117/1.3662089
Published in: Journal of Applied Remote Sensing Volume 5, Issue 1
Show Author Affiliations
Henrique G. Momm, The Univ. of Mississippi (United States)
Greg Easson, The Univ. of Mississippi (United States)


© SPIE. Terms of Use
Back to Top