Hyperdimensional data exploitation through parametric reduction

Feature reduction of hyperspectral data is a big challenge, particularly because the reduced dimensions must preserve the separability properties and key information content. Nevertheless, various techniques have been developed so far and are well documented in the literature. Here we characterize a novel technique of feature reduction, with main emphasis on the ability of enhancing the informative content of the reduced dataset, for data exploitation purposes. The parametric reduction of hyperspaces using the Exponential Gaussian Optimization (EGO) approach allows the analyst to quickly explore the dataset in terms of the occurrence and properties of the diagnostic features and the local albedo, as well. As a consequence, this technique is able to provide new insights into the accomplishment of the delicate task of hyperspectral classification.