Signal Subspace Estimation in Hyperspectral Data for Target Detection Applications

Dimensionality Reduction (DR) is a crucial first step in many hyperspectral processing algorithms. In some applications, such as target detection, change detection and classification, it is important to preserve the information associated to rare pixels, i.e. pixels scarcely represented in the data and containing spectral components that are linearly independent of the background. This paper presents a new method, unsupervised and fully automatic (i.e., it does not depend on tuning parameters), to estimate the signal subspace addressing both the abundant and the rare vectors subspaces. Experimental results demonstrate that the proposed algorithm outperforms other state of the art algorithms and hence provides effective new options for dimensionality reduction of hyperspectral remote sensing imagery.