Research On Hyperspectral Image Restoration Algorithm Based On Low-rank Tensor Restoration

Hyperspectral image remote sensing technology is a technology that integrates spatial two-dimensional image information and spectral dimensional information.At present,it has a wide range of applications in many fields,such as resource detection,landform information monitoring,ecological research and so on.However,due to various reasons,hyperspectral images often suffer from various noise interference or image degradation in the process of acquisition and transmission.It is necessary to restore them to improve the accuracy of subsequent applications and meet the actual engineering needs.Low-rank decomposition restoration algorithm using the low-rank property of hyperspectral images is a common restoration algorithm for hyperspectral images.However,due to the lack of prior knowledge,the restoration results of these methods are limited.In this thesis,we explore the problems existing in the above hyperspectral image restoration algorithms,and the main work is as follows:(1)A hyperspectral image restoration method based on local smoothness low-rank decomposition is proposed.In this method,the local smoothness of the image is also taken into account as prior knowledge,the gradient information of three dimensions is used to replace the original data information of the image for low-rank decomposition,and the influence factors are added for different dimensions to make the constraint penalty of the spatial and spectral components of the regularization term more flexible.The combination of global low-rank regularization and total variation regularization also enhances the image restoration effect.In addition,the non-convex exponential function(ETP)is used to replace the nuclear norm constraint of the matrix to further enhance the recovery ability of the low-rank matrix.On this basis,the Augmented Lagrangian Function(ALM)is used to optimize the model.The experimental results show that the proposed algorithm has a good effect in the process of hyperspectral image restoration,and can effectively remove the Gaussian noise and low-rank noise suffered by hyperspectral images.(2)This thesis proposes a hyperspectral image restoration method based on spatial-spectral consistency low-rank decomposition.By adding an additional spatial-spectral consistency regularization term,the data recovery ability of the low-rank model is improved.In this chapter,the existence and uniqueness of the solution of the model are proved,and a multi-block alternating direction method of Multipliers(ADMM)is used to optimize the model.The final experimental results show that the proposed algorithm can denoise and complete the hyperspectral image data after Gaussian noise,speckle noise,dead line noise and other types of noise,and achieve good results,and improve the performance of hyperspectral image data restoration.