Hyperspectral Image Low-Rank Recovery Based On Spectral-Spatial Weighted Total Variation

Recently,hyperspectral image(HSI),which has rich information in space,ra-diation and spectrum,has been widely used in environmental monitoring,precision agriculture,mineral recognition,military surveillance and other areas.However,due to various factors in the process of obtaining and transmitting,HSI will be blurred,noised and so on,which will seriously affect the accuracy of the subsequent HSI classifi-cation and recognition etc.HSI low-rank recovery with weighted spectral-spatial total variation is studied in the paper,and the proposed methods have important theoretical significance and application values.The main works of this paper are summarized as follows:(1)Based on local neighborhood weighted spectral-spatial total variation reg-ularization,a HSI low-rank restoration model is proposed.Firstly,by combing the spectral-spatial total variation and the low-rank prior,a HSI restoration model is proposed.In the model,spatial structure and spectral correlation of HSI are both considered.Secondly,a spectral-spatial total variation is improved,in which the spa-tial structure information is fully exploited.Then a new local spatial neighborhood weighted spectral-spatial total variation regularization model is developed.Finally,a fast algorithm based on alternating direction multiplier method is designed,and the parameters are discussed to analyze the stability of the method.The experimental results show that the proposed methods can retain the spatial structural information and spectral information well while removing the mixed noise.(2)A HSI restoration model based on non-convex low-rank relaxation and weight-ed spectral-spatial total variation is proposed.The low-rank term in the model men-tioned above,leads to the instability of the numerical solution.To improve the stabil-ity,matrix γ-norm is introduced to approximate the low-rank prior term of the matrix in the model.Then a new HSI non-convex low-rank relaxation restoration model is proposed,and the non-convex problem is solved in the framework of alternating direc-tion multiplier method.A large number of data experiments show that the proposed method can effectively remove the mixed noise,and has better robustness and stability.