Unsupervised Spatial Feature and Change Detection in RS Imaging

摘要

This is a conclusive report. Pending possible further funding it was decided to continue further R&D development of the system within the limited means of the University of Amsterdam with the UNSUP software as is. Target is to experiment with the UNSUP software in its present state (see previous report) with data sets in various user contexts, in order to determine the most promising lines for completing this promising state-of-the art module of unsupervised classification of multispectral remote sensing images. We investigated other related European studies to which we had access. It suggested a focus on the interface of RS (rasterized) to GIS (vectorized) processing and analysis, i.e. best practice method for mapping unsupervised class features in RS imagery to land cover classes for an area; effective mapping of signature based land cover classes to best fitting (local) geo-administrative land use classification for that area; associated methods of change detection on short/ long term intervals; associated methods of disaster residuals or pollution detection. We shall exploit the competitive edge of UNSUP by applied experiments with UNSUP together with other users in the US or Europe. The emphasis of this research will be on the applied methodology in the context of prevailing GIS processing environments (interfaces to GIS packages, S-Plus environment, Arcinfo). This might best pave the way for final funding to achieve the system for general use.