Generation of land-cover descriptions for the classification of multisensor remote-sensing images

The paper deals with a learning system to acquire automatically the descriptions of objects to be recognized in remotely sensed images. These descriptions are in a symbolic form and use contextual properties and fuzzy logic descriptors to describe spectral properties. The system was tested on an Airborn Thematic Mapper (ATM) image of agricultural crops. The satisfactory results obtained in terms of classification accuracy witnessed the goodness of the approach and that the acquired descriptions are reliable. In particular, they can be employed as knowledge base for an expert classification system.<
>