Spectral Unmixing based on Joint Sparsity and Total Variation using Remote Sensing Data

Abstract

Hyper-spectral imaging is part of a process known as spectral imaging. The aim of hyperspectral imaging is to find the spectrum in the image of a scene for each pixel. Hyperspectral unmixing in hyperspectric image analysis is emerging as a subject in which the materials contained in an image and therefore the share of each material found in an image can be separated. The different materials are referred to as end members and proportion values as maps of abundance. Hyperspectral unmixing is an important method for estimating fractions of different types of terrestrial coverage from remote sensor images. It calculates the fractional abundance of the components present in every pixel of hyperspectral images. Multiple noise types also influence a hyper-spectrum image in a general situation where mixed noise is present, joint sparsity and total variance (JSTV) deals with hyperspectral unmixation. The common sparsity was developed to take advantage of the abundance maps. In order to model abundance maps, a total variation based regulatory structure was also used. The SplitBregman technique was used for the solution of an optimization problem using an algorithm. The results show that the proposed joint sparsity and complete methods of variations are able to unmix real hyperspectral and synthetic data, while retaining spatial endpoints with smooth maps.