Satellite image classification consists in assigning each pixel of an image to a particular class:

• vegetation,
• crops,
• rivers, ...

Use of all the available information helps to make the best decision to improve classification.

In remote sensing, several kinds of images can be used to increase the knowledge about each pixel, such as

• spectral bands of a satellite image and
• out-image data describing the geographical context of the studied area:
• road maps, river maps,
• town locations,
• elevations,
• slope orientations and so on

Information fusion is the operation consisting in aggregating all the available information.

In the following pages, the data fusion operators of

• probability theory (Bayes' rule),
• belief theory (Dempster-Shafer's orthogonal rule) and
• possibility theory (T-norms, adaptive operators, priority operators)

are presented.

The difficulties arising from the adaptation of these operators to a LANDSAT satellite image classification problem are considered. Finally, a comparative study of the different operators is presented.