Feature Extraction-Based Hyperspectral Unmixing

Abstract

Purpose Hyperspectral imaging belongs to a class of techniques called spectral imaging or spectral analysis. Due to the high dimensionality of hyperspectral cubes, it is a very difficult task to select few informative bands from original hyperspectral remote sensing images. The dimensionality reduction of hyperspectral images is a pre-processing technique used to perform many applications like unmixing, classification, reconstruction and detection. Procedures Hyperspectral unmixing is an emerging topic in hyperspectral image analysis to distinguish the materials present in an image and thereby finding the proportion of each material in an image. The distinct materials are called as end members or spectral signatures, and proportion values are called as abundance fractions. This paper proposes a scale invariant feature transform (SIFT)-based dimension reduction with application to unmixing pixels in hyperspectral images. Results The proposed feature extraction-based selection of non-redundant informative bands followed by unmixing of pixels has been proved qualitatively and quantitatively with comparison to existing techniques principal component analysis, linear discriminant analysis-based unmixing. Another important advantage of the proposed method is that it takes into account the spectral variability in materials. Conclusion The proposed technique has been highlighted using the performance measures spectral angle distance and abundance angle distance.