ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume III-7
https://doi.org/10.5194/isprs-annals-III-7-17-2016
https://doi.org/10.5194/isprs-annals-III-7-17-2016
07 Jun 2016
 | 07 Jun 2016

VIRTUAL DIMENSIONALITY ESTIMATION IN HYPERSPECTRAL IMAGERY BASED ON UNSUPERVISED FEATURE SELECTION

M. Ghamary Asl and B. Mojaradi

Keywords: Hyperspectral Imagery, Unsupervised Feature Selection, Signal Subspace Identification, Virtual Dimensionality, Partition Space

Abstract. Virtual Dimensionality (VD) is a concept developed to estimate the number of distinct spectral signatures in hyperspectral imagery. Intuitively, detecting the number of spectrally distinct signatures depends on determining the number of distinct bands of the data. Considering this idea, the current paper aims at estimating the VD based on finding independent bands in the image partition space. Eventually, the number of independent selected bands is accepted as the VD estimate. The proposed method is automatic and distribution-free. In addition, no tuning parameters and noise estimation processes are needed. This method is compared with three well-known VD estimation methods using synthetic and real datasets. Experimental results show high speed and reliability in the performance of the proposed method.