Fusion of methods for the classification of remote sensing images from Urban areas

The fusion of methods for the classification of panchromatic high resolution satellite remote sensing images from urban areas is addressed. In this paper we propose to aggregate the results provided by different classifiers with complementary properties. Classical combination rules fail to properly combine conflictual information or information provided by sources with different reliabilities. To overcome those problems, the proposed approach is based on the definition of two measures of accuracy. Based on fuzzy set theory, both a local and a global accuracy are defined for each classifier. The fusion is then performed with an adaptive fuzzy combination operator. In terms of classification accuracy, the proposed method performs better than each classifier used separately. Results are presented on IKONOS images.