Pixels of a satellite image must be classified into several classes of vegetation. To classify a pixel means to store it in the class which its profile is closest. Resemblance is a vagueness concept, it is thus necessary to establish the particular features of each class, allowing to distinguish them to each other. That is denoted the spectral signature of a class.

Knowing the signature of each class, the classification step is carried out: the pixels of each image are classified in the class appearing to be most plausible. Then, this classification is performed for any image. During the classification step, each image brings information on its own pixels. Such an information is used to carry out classification.

Methods of fusion of information are used to share all the information given by each source on the classification of pixels. This reunion of information allows a total classification, surer and more accurate than separate classifications of each source. So the performances of the classification performed by using these various methods are presented.