Satellite images are numerous and weakly exploited: it is urgent to develop an efficient and fast indexing/retrieval system to easy their access. Content-based image retrieval systems (CBIR) are known to provide an efficient framework. We thus propose to associate a CBIR approach with text-based queries to adapt to these big (12000$\times$12000 pixels) and semantically rich images. The presented system relies on a multimedia data mining system called PLATO able to adapt to any kind of multimedia data. Moreover state-of-the-art relevance feedback strategy is introduced to provide interactive learning and auto annotation.