Urban areas classification tests using high resolution pan-sharpened satellite images

Abstract

The possibility of transferring high spectral contents of medium geometric resolution images obtained from traditional satellite images (TM; ETM+) and newer ones (ASTER, ENVISAT) to high resolution images has been considered to resolve the problems connected to large scale classification. This work suggests an operational approach to this problem. It points out that the aspects related to obtaining good results in an easy, economic and rapid way are as important as the scientific and technological aspects. The suggested method is based on the well known pan-sharpening technique; only a limited amount of experience can however be found in literature concerning its verification for real applications. The authors do not intend proposing new pan-sharpening algorithms in this paper, but rather to demonstrate how its correct use and the customisation of already known techniques (mainly used for aesthetic purposes) can produce interesting scientific results and can also solve some practical problems such as the management of large size data. In what follows that following is illustrated: the techniques that were adopted to generate pan-sharpened synthetic bands; some radiometric verifications that were performed on Landsat 5 TM are shown as are the results of elaborations on QuickBird images; some results of LVQ neural classifications that were carried out on 4 bands of a QuickBird image in an urban area generated with the previously described technique. A preliminary qualitative analysis has shown how a classical pixel-based classification approach, such as the one that is here proposed, is not sufficient to generate suitable thematic images of the correct discrimination of urban environments.