Abstract: A break through development in remote sensing is Hyperspectral Imaging. Imaging Spectrometers, often referred to as hyperspectral cameras (HSCs) are used for hyperspectral imaging and they acquire images with higher spectral resolution than multispectral cameras. Due to low spatial resolution of HSCs, spectra measured by HSCs are mixtures of spectra of materials in a scene and each pixel is assumed to be a mixture of few materials, called endmembers. This necessitates unmixing which involves estimating the number of endmembers, their spectral signatures and their abundances at each pixel. Various algorithms like HYSIME, VCA, DECA, NMF, N-Finder were introduced for hyperspectral unmixing. A sparse regression scheme based on compressive sensing is also used for identifying pure form of pixels of a scene. This approach reduces the number of endmembers needed to represent the data and provides more robust solutions. A collaborative Sparse Regression method is also developed which can be implemented in parallel nature and thus improves the speed of operation and accuracy. In this paper a brief study of various unmixing algorithms were presented along with a comparison of their performance.

Keywords: Hyperspectral Imaging, Hyperspectral Unmixing (HU), endmembers, Compressive sensing, Hysime, VCA, DECA, Spectral library.