In this paper we propose a hardware architecture of the Fuzzy ART's generalization applied to the classification of multi-spectral image acquired by the Algerian satellite ALSAT 1. The developed solution is original and exploits the maximum benefit of the embedded resources in the reconfigurable circuits FPGAs. The goal is the research of a scalable architecture which gives a very good quality of classification and a very weak computing time (about a microsecond) with a very limited occupation of the logical resources. This architecture is easily reconfigurable.