Hyperspectral Image Unmixing Algorithm And Software System Based On Space Spectrum Joint Prior

With the rapid development of hyperspectral imaging technology,hyperspectral images have been widely used in environmental monitoring,target detection,mineral exploration and so on.However,due to the low spatial resolution of hyperspectral imaging,pixels in hyperspectral images are not pure pixels,while usually contain mixed materials for each pixel.Therefore,it is necessary to unmix the hyperspectral data to extract both pure materials(endmembers)contained in the scene and their corresponding content coefficients(abundances).As a blind source separation problem,Hyperspectral unmixing is usually an ill-posed problem due to insufficient information.In order to obtain more prior knowledge as much as possible,the goal of the paper is to investigate the spectrum-space constrained nonnegative matrix factorization(NMF)models to obtain more accurate results.And the models are established by using spatial contexts and spectral prior information extracted from the scene.In this paper,after investigating the spectral constrained NMF unmixing methods,we further propose two spectrum-spatial constrained NMF unmixing methods.The main contributions of this thesis are listed in the following.1)Traditional spectral information constrained NMF based hyperspectral unmixing algorithms are studied.We make discussions on the mechanism of NMF based unmixing models and summarize their advantages and disadvantages.Furthermore,a comprehensive assessment are made to illustrate the performance for these unmixing algorithms.2)A spatial gradient sparsity constrained NMF(SGSCNMF)algorithm is proposed,and it utilizes the piecewise smoothing of the hyperspectral image,the geometrical properties of the endmembers,and the sparsity of abundances in the hyperspectral image.Under the NMF framework,we present a jointly spectral-spatial optimizing model for the unmixing problem.The proposed model benefits from three aspects:1)the spatial gradient sparsity constrained regularization of abundances,2)the minimum distance constrained regularization of endmembers,and 3)the?_???_?constrained regularization of abundances.Finally,the model is solved by Alternating Direction Method of Multipliers(ADMM).Experimental results on synthetic dataset and real data demonstrate that the proposed algorithm outperforms the spectrum only constrained methods.3)Considering the spatial correlations of hyperspectral image,we firstly present a new spatially weighted sparsity regularization term for abundance map.Under the NMF unmixing framework,we then propose a multi-constrained NMF model by incorporating the spatially weighted sparsity regularization term of abundance,and the geometrical constrained term of endmember.Finally,the ADMM is used to solve the minimizing model.The proposed method is validated on both simulated and experimentally measured data,which is proved that this method can achieve better unmixing results compared with several state-of-the art algorithms.4)A hyperspectral image unmixing software system is developed by integrating six algorithms described in this paper.Specifically,it contains three core modules:unmixing module,data analysis module,and visualization module.