DIMENSIONALITY REDUCTIONTECHNIQUES FORANALYSIS OF HYPERSPECTRAL DATA

A.Kiranmai, Y.Sai Praveen, Iyyanki. V. Murali Krishna, V.Sowmya Devi

Abstract: The Hyperspectral images provide the images with hundreds of narrow contiguous spectral channels. The spectral information provided by the hyperspectral images is high when compared to the other class of remote sensing images such as panchromatic, multispectral images.Even though the hyperspectral images containsufficient spectral information but processing and exploiting the hyperspectral data is considered as challenging task because of its high dimensionality. Data redundancy is one of the problems when processing hyperspectral data. And some bands in hyperspectral images are noisy and some bands are set to zero. So when we use all the bands for processing it increases the computational complexity in terms of storage and processing time. Because of this problem Dimensionality reduction techniques must be applied before we use it for processing. With that we can map high dimensionality of hyperspectral image data to lower dimensions without losing much of spectral features provided by the original hyperspectral data cube. Many techniques are developed for this dimensionality reduction; three of those techniques are analyzed in this paper.

Keywords: Dimensionality Reduction, Hyperspectral, Band Selection, Principal Component Analysis,Minimum Noise Fraction, Independent Component Analysis

DOI: https://doi.org/10.15623/ijret.2016.0519002