Research On Nonlinear Hyperspectral Unmixing Algorithm Based On Multilinear Mixing Model

Spectral unmixing of hyperspectral images is an important issue in the fields of remote sensing.Jointly exploring the spectral and spatial information embedded in the data is helpful to enhance the consistency between mixing/unmixing models and real scenarios.In this thesis,we first propose a graph regularized nonlinear unmixing method based on the recent multilinear mixing model（MLM）.The MLM takes account of all orders of interactions between endmembers,and indicates the pixel-wise nonlinearity with a scalar probability parameter.The proposed algorithm makes full use of the underlying manifold structure of the pixel spectrum by adding Laplacian regularization terms,which improves the estimation of abundance and nonlinear probability parameters.In addition to the spectral-based regularization term,abundance sparse constraint is added to the model.The resulting optimization problem is addressed by using the alternating direction method of multipliers（ADMM）,yielding the so-called graph regularized MLM（G-MLM）algorithm.To implement the proposed method on large hyperspectral images in real scenarios,we propose to utilize a superpixel construction approach before unmixing,and then apply G-MLM on each superpixel.The proposed methods achieve superior unmixing performances to state-of-the-art strategies in terms of both abundance and probability parameters,on both synthetic and real datasets.The noise in the real hyperspectral data will seriously degrade the performance of various unmixing algorithms,so enhancing the robustness of the unmixing algorithm is a subject worth studying.This thesis also proposes a robust MLM（R-MLM）,which uses the robust l_2,1 norm as the loss function in the unmixing model,in order to reduce the interference of noise to the unmixing algorithm.To further improve the parameter estimation,we also consider to introduce the abundance sparse constraint to the model.The ADMM algorithm is applied to solve the resulting optimization problem.On both the synthetic and real hyperspectral datasets containing noisy channels,the proposed R-MLM yields better unmixing result compared with several state-of-the-art methods.