BasedonDecisionFusion.PDFVIP

A Combined Support Vector Machines Classi?cation Based on Decision Fusion Mathieu Fauvel ?, Jocelyn Chanussot ? Jon Atli Benediktsson  ?Signals Images Laboratory - LIS Grenoble Department of Electrical and Computer Engineering Grenoble National Polytechnical Institute - INPG University of Iceland BP 46 - 38402 St Martin d’Heres - FRANCE Hjardarhagi 2-6, 107 Reykjavik, ICELAND E-mail: {mathieu.fauvel, jocelyn.chanussot}@lis.inpg.fr E-mail: benedikt@hi.is Abstract— Decision fusion for classi?cation of hyperspectral data morphological feature extraction does not provide information from urban area is addressed. Classical classi?cation algorithms about the size, the shape or the orientation of the structure. For are based on the spectral signature of the individual classes. For an urban area context, both kinds of information are needed to urban area, where classes could be de?ned in accordance with the shape of the structure, these methods have a major drawback: no allow a ?ne classi?cation. spatial information are contained in the spectrum. A new method In this paper, we propose to fuse the results obtained by a has been proposed that considers the spatial content, but it reduces separate use of the spectral data and the Extended Morphological the spectrum to a small number of bands and does not exploit the Pro?le. Each data are processed by SVMs classi?ers. The SVMs spectral richness of the hyperspectral data. In this paper, we propose where chosen according to their strong capability to deal with to use both approaches, and then fuse them. The data are ?rst pre- processed to extract some spatial information. Using Support Vector remote sensing data [3], [4]. The result from each classi?ers are Machines (SVMs), the data are classi?ed. Finally, according to the aggregated according to the intrinsic characteristic of the SVMs