Comparing pixel vs. object based classifiers for land cover mapping with Envisat- MERIS data

Abstract

The work presented in this paper is part of SatStat project, which is being developed in e-Geo – Geography and Regional Planning Research Centre of the Universidade Nova de Lisboa, under the framework of the European Space Agency (ESA) Announce of Opportunities for Portugal. The main goal of SatStat is to annual monitor forest areas using low resolution images. The imagery dataset includes an Envisat-MERIS multitemporal set of images for Portugal. In order to monitor the forest areas, 2003 was consider as the reference year and several classification techniques were tested to map the land cover of Portugal in that year, using a 2 level nomenclature. A pixel-based classifier was tested against an object-oriented classifier and an accuracy assessment was preformed to identify the best method.