Landscape Change Detection from Multispectral Image Data Using Hyperclustering
by
Wayne Myers
Environmental Resources Research Institute, Pennsylvania State University, USA

Hyperclustering of multiband (multivariate) image data serves to segment the scene (landscape) according to prominent spectral patterns of the environment. The number of such clusters is large, being chosen a priori to saturate the informational capacity of a byte of computer storage. Hyperclustering is informationally advantageous for detecting changes from image datasets collected at different times over an area. It reduces the volume of data to be processed in change detection and filters noise from conventional change indicators. The spatial segmentation further enables new kinds of change indicators involving comparison of spatial organization between temporal image datasets. The spatial comparisons can be accomplished either by spectral averaging or spectral matching. The comparisons are temporally directional in the sense that going from earlier to later may reveal different aspects of change than retrospective comparison. Hyperclustering strategies can also be extended to analysis of changes spanning more than two occasions of image data collection. The multi-temporal datasets are first hyperclustered individually, and change indicators developed by two-date comparisons. The two-date change indicators are then interleaved as if they were spectral bands in a multiband image dataset. Hyperclustering is applied to the (multiband) composite of change indicators, with the resulting clusters serving to segregate different types of temporal change phenomena.

Date received: October 9, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cafr-17.