Possibility theory for supervised classification of remotely sensed images: A study case in an urban area in Algeria

In this paper we present a possibilistic classifier of multispectral remotely sensed images. This classifier developed in the framework of possibility theory is based on a fusion process using several kinds of combination operators (conjunctive and disjunctive). Unlike the probabilistic classifier which can model only the data uncertainty through a probability measure, the possibilistic classifier has the ability to handle both uncertainty and imprecision of pixel classification through a possibility and a necessity measures. These two classifiers are applied to classify a multispectral image acquired on 2001 by ETM+ sensor of Landsat-7 satellite. This multi-band image covers a north-eastern part of Algiers (Algeria). Compared with probabilistic classifier, the possibilistic one is advantageous in reducing the error and confusion between the different classes. Indeed, the statistical assessment of possibilistic result indicates that the overall accuracy is improved from 72.72% to 90.23% and Kappa indicator increases from 0.62 to 0.86.