Unsupervised feature selection applied to SPOT5 satellite images indexing


Marine Campedel, Ivan Kyrgyzov, Henri Maitre ;
Proceedings of the Workshop on New Challenges for Feature Selection in Data Mining and Knowledge Discovery at ECML/PKDD 2008, PMLR 4:48-59, 2008.


Satellite images are numerous and weakly exploited: it is urgent to develop efficient and fast indexing algorithms to facilitate their access. In order to determinate the best features to be extracted, we propose a methodology based on automatic feature selection algorithms, applied unsupervisingly on a strongly redundant features set. In this article we also demonstrate the usefulness of consensus clustering as a feature selection algorithm, allowing selected number of features estimation and exploration facilities. The efficiency of our approach is demonstrated on SPOT5 images.

Related Material