Removing ambiguities in a multispectral image classification

This article deals with the processing of already classified satellite images according to land use in order to remove ambiguities, i.e. mistakes in labels. Those images have already been classified with the maximum likehood method but some classes are not correctly determined. For the elimination of ambiguities in this kind of class, the authors applied their method of determination of land use mixture in pixels. They first briefly review their method of determination of land use mixture. Then they explain how they deal with ambiguities in labels of the maximum likehood classification. They finish with three examples of satellite images that have not correctly been classified. The first one is the vineyard case. Another example for bare soil and urban zone. The last one is a forestry survey application, the determination of the planted pines density.