Hyperspectral Image Unmixing Based On Local Joint-Sparse

In the past decades,remote sensing imaging technology has been widely developed.However,due to the limitations of hardware devices and the natural variability,there are a large number of mixed pixels in hyperspectral images.The purpose of hyperspectral image unmixing is to obtain components(endmembers)and corresponding proportions(abundance)from mixed pixels.Assuming a known spectral dictionary,exploration of spatial and spectral information helps to obtain accurate abundances.In recent years,spectral unmixing based spatial information has earned a lot of interest.Based on blockjoint-sparse unmixing method first partitions the abundance matrix,and then considers the joint sparsity between adjacent pixels in the abundance matrix.Bilateral block joint sparse structure,taking into account both horizontal and vertical block joint sparsity,has achieved good results in unmixing.However,since the original hyperspectral image is 3-D,expanding it into a matrix causes the loss of spatial information.Although it is effective to consider the unfolding ways of abundance matrix with both directions,fully utilizing the spectral and spatial information is still difficult in hyperspectral data.This thesis first proposes a local-sparse structure on similar pixels for hyperspectral unmixing and construct a selection matrix for clustering similar pixels.Specifically,we assume that local similar pixels share almost identical endmembers.In this vein,we focus on utilizing local spatial detail information of hyperspectral data and enhance the sparsity of local related pixels.Moreover,combined with a low-rank property,we propose a local spatial similarity based joint-sparse regression unmixing model and derive an algorithm under the alternating direction method of multipliers framework.This thesis tests and compares the new algorithm with several classical unmixing algorithms on two simulated datasets and two real datasets.Experiments show that the new algorithm is significantly superior to other methods in terms of numerical results and abundance map estimation.