Hyperspectral Image Denoising Algorithm Based On Low Rank Description Model

Hyperspectral images have been widely used in military reconnaissance,urban spatial planning,ocean monitoring and other fields because of their rich spectral and spatial information,which can accurately distinguish landform information.However,in the acquisition process of hyperspectral imaging,the image is often degraded due to the pollution of various noises,which ultimately affects the visual quality of the image.Therefore,denoising of hyperspectral images is a key task.With the development of low-rank theory and image restoration techniques,researchers have applied traditional low-rank and total variational methods to spectral image denoising according to the characteristics of hyperspectral images.However,the existing low-rank model methods cannot well mine the relationship between the spectral image spatial domain and the spectral dimension,and cannot effectively remove the mixed noise in the image.Aiming at the problem of noise removal of hyperspectral images,based on the low-rank description model,this paper makes full use of the intrinsic spatial information between adjacent bands of spectral images,aiming to improve the restoration quality of hyperspectral images.The main research work of this paper is as follows:(1)A hyperspectral image denoising method based on low-rank restoration and spatial spectral total variation is studied.Using the low-rank and spatial characteristics between adjacent bands of the image,the traditional low-rank matrix method is used to suppress the separation sparse noise,the spatial constraints are introduced to remove the structured sparse noise,and the spectral total variation is used to enhance the correlation of the global spatial spectrum.Finally,the mixed noise in the image is removed,and the restoration accuracy of the image is improved.(2)A hyperspectral image denoising method based on tensor decomposition and total variation constraint is studied.This paper takes the tensor method as the research background,uses the lowrank tensor decomposition to separate the noise from the original image,and adds an improved weighted total variation regularizer to the low-rank decomposition framework to fully characterize the hyperspectral image band.The spatial and spectral correlations between and the smoothness of the segments are normalized by the L1 norm and the F norm to normalize sparse and Gaussian noise,respectively.The tensor-based method basically preserves the non-local similarity and spectral correlation of the image spectral space,and has better restoration results.(3)A hyperspectral image denoising method based on spectral low-rank and spatial non-local selfsimilarity is studied.This paper takes the low-rank characteristics of hyperspectral image spectra as the research premise,introduces low-rank-based non-local self-similarity into the sparse representation model to represent spatial features,regularizes the dictionary,and fully exploits the spatial information of spectral images.to separate the noise part to further remove noise and improve the denoising ability of the image.Compared with other algorithms,the experimental results show that the peak signal-to-noise ratio and structural similarity are significantly improved.Experiments are carried out on a variety of hyperspectral datasets,and the appeal denoising method is evaluated by the peak signal-to-noise ratio PSNR,structural similarity SSIM and the erreur relative globale adimensionnelle de synthese ERGAS image quality evaluation standard.The experimental results show that the proposed method is better than Other methods have better denoising effect on hyperspectral images.