Research On Algorithms Of Hyperspectral Remote Sensing Images Unmixing Via Spatial Spectral Information

Spatial information is increasingly becoming vital supplementary information in the field of hyperspectral endmember extraction since it takes into consideration the spatial correlation of pixels,which generally involves jointing spectral information for preprocessing and/or endmember extraction in hyperspectral imagery.Endmember extraction algorithms(EEAs)are one of the most widely discussed technique in the field of hyper spectral processing.Most of spectral-based EEAs exploit the convexity of the data structure in the spectral domain,yet fail to take the spatial attribute of the endmembers into account.Spatial-spectral-based EEAs focus on the discussion of a combination of spatial correlations and spectral features between pixels for identifying endmembers,which require a rigorous tuning process for parameters to optimize endmember extraction performance at a price of the computational burden.Preprocessing is a major area of interest within the field of hyperspectral endmember extraction which generally involves the process of utilizing both spatial contextual and spectral features with the specific intent of offering a few high-quality pixels for fast-and accuracy-orientated endmember extraction.However,most of PPAs use time-consuming clustering methods,such as k-means,fuzzy c-means,or sliding window,to capture spatial information.They also rely on parameters or thresholds to maintain preferable endmember candidates,which requires manual tuning depending on data quality measures such as noise or image homogeneity,and thus are difficult to understand intuitively.For the aforesaid flaws of existing PPA and EEA,this thesis presents a new preprocessing algorithm based on superpixel segmentation,box theory,and pure pixel index,and also presents two endmember extraction algorithms based on spatial weight,data convexity,and spatial energy prior,the three algorithms are described as follows:1.This thesis firstly proposes a superpixel-guided preprocessing algorithm based on spatialspectral analysis,named SGPP.The proposed SGPP algorithm first uses a well-known superpixel method,i.e.,the simple linear iterative clustering(SLIC)algorithm,to segment the whole image into a set of superpixels.For the pixels originating from the superpixel,the SGPP then captures their spatial compactness and spectral purity based on the box theory and pure pixel index.The SGPP finally retains a few high-quality pixels from each superpixel with high spatial compactness and spectral purity index.The experimental results indicate that SGPP can reduce the computational burden of spectral-based EEA,and meanwhile retain endmember accuracy.2.Then,this thesis proposes a spatially weighted simplex strategy,called SWSS,for hyperspectral endmember extraction that investigates a novel integration framework of the spatial information-embedded simplex for identifying endmember.Specifically,the SWSS generates the spatial weight scalar of each pixel by determining its corresponding spatial neighborhood correlations for weighting itself within the simplex framework to regularize the selection of the endmembers.The SWSS could be implemented in the traditional simplex-based EEAs,such as vertex component analysis(VCA),to introduce spatial information into the data simplex framework without the computational complexity excessively increasing or endmember extraction accuracy loss.3.Finally,this thesis develops a spatial energy prior constrained maximum simplex volume approach,called SENMAV,for spatial-spectral endmember extraction of hyperspectral images(HSIs).Our proposed SENMAV method investigates the spatial information from the perspective of spatial energy prior of Markov random field(MRF),and the spatial energy prior is utilized to be as a regularization term of the traditional maximum volume simplex model to constrain the selection of the endmembers in both spatial and spectral viewpoints simultaneously.This algorithm sheds new light on spatial-spectral-based EEAs that the proposed SENMAV algorithm well balances the trade-off between the endmember extraction accuracy and spatial attribute requirement of endmembers since SENMAV jointly take into consideration the data convexity and spatial energy prior of endmembers.To sum up,this thesis proposes three algorithms;SGPP introduces spatial information in the preprocessing stage to select a high-quality data subset for subsequent endmember extraction;SWSS considers the defect of simplex framework and introduces spatial weight scale to regularize traditional simplex for the for purpose of endmember optimization;SENMAV directly introduces spatial energy to improve the endmember extraction capacity of traditional simplex.The experimental results indicate that the proposed algorithms have obvious improvements over other EEAs.