Research On Hyperspectral Image Denoising Algorithm Based On Total Variation And Low Rank Characteristics

Hyperspectral images carry both ground object information and spectral information,which can better discriminate and classify ground objects,and have gained extensive attention in fields such as military reconnaissance,agricultural production,and target detection.However,a variety of noises are often generated in the process of imaging and transmission,such as Gaussian noise,salt and pepper noise,dead lines and stripes,etc.The existence of noise will not only reduce the image quality,but also affect the accuracy of subsequent work.Therefore,as a preprocessing step in the application of hyperspectral image data,denoising the hyperspectral image has important academic significance and research value.This thesis will focus on the research on hyperspectral image denoising algorithm.The main work is as follows:In the process of writing the thesis,the author has read a lot of literature,and analyzed the research process of hyperspectral image denoising algorithm,and also studied the hyperspectral image denoising algorithm based on total variation(TV),low-rank matrix,and tensor.Finally,the advantages and disadvantages of the existing algorithms are summarized.The existing hyperspectral image denoising algorithms fail to fully explore the structural information and detail information of the image,making the resulting image too smooth,and the denoising performance is poor in the case of high Gaussian noise intensity.Aiming at the above problems,a new hyperspectral image denoising algorithm is proposed,hyperspectral image hybrid noise denoising algorithm based on central difference total variation and local low-rank properties(LLCDTV).The algorithm constrains Gaussian noise and proposes a central difference total variation model(CDTV),which is able to explore image structural information.Simulation experiments show that the algorithm can remove the hybrid noise in hyperspectral images.For the hyperspectral image hybrid noise denoising algorithm based on central difference total variation and local low-rank properties,the central difference total variation model is more sensitive to noise,which leads to the weaker robustness of the algorithm.A new hyperspectral image denoising algorithm is proposed,hyperspectral image hybrid noise denoising algorithm based on spatial center difference total variation and band fusion strategy(SCDTVBF).The algorithm optimizes the central difference total variation model into a spatial central difference total variation model(SCDTV),and proposes a band fusion strategy for two-stage denoising.Simulation experiments show that the algorithm overcomes the shortcomings of the hybrid noise denoising algorithm for the hyperspectral image hybrid noise denoising algorithm based on central difference total variation and local low-rank properties,and shows better denoising results.The existing hyperspectral image denoising methods based on low-rank characteristics still have common problem,that is,image details are easily lost,and the algorithm is less robust.And the algorithm based on tensor spends longer time.Aiming at the above problems,a new hyperspectral image denoising algorithm is proposed,low-rank tensor hyperspectral image weighted hybrid noise denoising algorithm based on space-spectral factor regularization(WSSFR).The algorithm combines the low-rank tensor denoising model with the Gaussian noise constraint model,and proposes a denoising model with good denoising effect and robustness.Simulation experiments show that the algorithm overcomes the shortcomings of the existing hyperspectral image denoising methods based on low-rank matrix restoration,and can better separate the image signal from the noise.