Researches On Hyperspectral Image Unmixing Based On Maniflod Constraint

The hyperspectral image contains high spectral resolution with tens to hundreds of spectral bands which is applied in many fileds,such as agriculture,the environment,surveying,and so on.However,due to the low spatial resolution of hyperspectral images,different types of materials may exist in the same pixel of a hyperspectral image.The mixed pixels problem not only influences the precision of object recognition and classification,but also becomes an obstacle to quantification analysis of remote sensing images.Hence,the hypesspectral unmixing(HU)technology emerges as the times require.Hyperspectral unmixing is to decompose the mixed pixels into dependent materials(called as endmembers)and their corresponding proportions(called as fraction abundances).Non-negative Matrix Factorization,(NMF)decompose a high-dimensional and positive data into the multiply of two lower-dimensional non-negative matrices.Recently,in the field of HU,NMF is one of hot topics because it has the advantage of approximating the linear spectral mixing model(LSMM)However,NMF cannot achieve a global minimum in general.HU based on NMF methods with various constraints is added to improve the accuracy of HU.This thesis based on NMF model,researches on the intrinsic manifold structure,build graph to model manifold containt to design several non-negative matrix decomposition of HU.Details are as follows:1.For existing manifold regularized NMF algorithm does not fully exploit the limitations of image spatial information,we propose a manifold regularized sparse NMF with superpixel for HU method(MRS-NMF).Based on each superpixel,an affinity graph is constructed to encode the spatial geometrical information and spectral geometrical manifold structure.Besides,the sparsity is integrated in the objective function,then the multiplicative iteration algorithm is used to optimize the solution.Experimental results demonstrate that the proposed method outperforms several the state-of-the-art methods.2.For most NMF methods ignore the geometric relationship of endmembers,we propose a endmember and abudance graph sparse constraint based NMF(EAGNMF)method for HU which integrates endmember sparse graph structure into NMF.The proposed algorithm considers the intrinsic manifold structure of both abundance and endmember for HU.In order to evaluate the effectiveness of our model,experiments have been performed on synthetic and real hyperspectral data sets,and the results show that the proposed method presents better performances when compared with several widely used unmixing methods.