Research On Hyperspectral Image Subpixel Mapping Methods

Due to the limitation of spatial resolution of spectral imaging technology and the complex diversity of ground distribution,there may be many different types of ground objects in a pixel,which is called a mixed pixel.The existence of mixed pixels is one of the main obstacles in the quantitative study of hyperspectral image.Hyperspectral image subpixel mapping technology is the key technology to solve the mixed pixel problem.In recent years,how to make full use of hyperspectral image spatial information and spectral information to solve the problem of subpixel mapping has attracted much attention.However,this kind of method still has problems such as large solving space and insufficient mapping accuracy.In view of the above problems,this thesis focuses on the hyperspectral subpixel mapping method on the basis of spatial-spectral.Our major work in this thesis is summarized as follows:1.The solution model for subpixel abundance has the problems of easily falling into local minima,we propose a constrained spatial-spectral subpixel mapping method.Firstly,according to the smoothness of the subpixel abundance matrix,anisotropic total variation is added into the subpixel abundance solving model to promote the smoothness of the subpixel abundance matrix.Secondly,the reweighted one-norm is added into the subpixel abundance solving model as a sparse regularization term to promote the sparsity of the subpixel abundance matrix according to the sparsity of the subpixel abundance matrix.Then,the recursion principle is used to solve the subpixel abundance.Then,the final subpixel mapping is performed by using the class determination strategy.Finally,two data sets are used for experiments.The results show that the proposed method can improve the accuracy of subpixel mapping.2.Aiming at how to solve the application problem of isotropic total variation in spatial-spectral subpixel mapping,a spatial-spectral subpixel mapping method based on isotropic total variation is proposed in the third chapter of this thesis.The isotropic total variation is applied as a constraint term to the subpixel abundance model.Isotropic total variation is a non-convex optimization problem,which is non-differentiable and difficult to solve.This thesis uses the Split Bregman algorithm to solve it.Compared with anisotropic total variational regularization constraints on two data sets,the results show that the proposed isotropic total variation constraints can improve the accuracy compared with anisotropic total variation constraints.3.Aiming at the spatial information loss problem of the spatial-spectral interpolation subpixel mapping method,an improved spatial-spectral interpolation subpixel mapping method is proposed in the fourth chapter.First of all,Due to the influence of spectral unmixing error,strictly following the abundance constraints to determine the number of subpixels in the mixed pixel for each class will result in the loss of spatial information,resulting in many isolated points.Therefore,an improved method which does not consider the abundance constraint in the process of class determination,but adopts the winner-take-all strategy is proposed.Finally,two data sets are used for experiments,and the results show that the proposed method can improve the accuracy of subpixel mapping.