Hyperspectral Image Denoising Based On Sparse Representation And Low Rank Model

With the development of remote sensing satellite technology,hyperspectral remote sensing images are widely used in geological prospecting,water quality evaluation,agricultural early warning,urban planning,disaster management,and military reconnaissance.Hyperspectral image denoising has played an important role in subsequent research in the field.With the development of low-rank sparse theory,more and more low-rank sparse models are used in hyperspectral image denoising tasks.However,the existing low-rank sparse algorithms cannot effectively represent the spatial and spectral features in hyperspectral images.Therefore,this article uses hyperspectral image denoising as the background and uses the inherent structural characteristics of hyperspectral images to develop low-rank sparse models.Research on the application of denoising.It aims to further improve the performance of sparse low-rank denoising and promote the practical application of hyperspectral images.The main research work of this thesis is as follows:1.A denoising algorithm based on graph structuring and three-dimensional non-local is proposed.The graph structured low-rank sparse model can not only describe the similarity between the bands in the hyperspectral image,but also maintain the distribution characteristics between the bands,ensuring the same spectral characteristics as the real hyperspectral image and reducing the spectral loss.The three-dimensional non-local constraint makes use of the three-dimensional spatial self-similarity in the hyperspectral image.It is an effective low-rank prior constraint and can improve the robustness of the algorithm.2.A denoising algorithm based on space-spectrum full variational constraints is proposed.Using the three-dimensional non-local similarity in the hyperspectral image to guide the solution of the local low-rank sparse model.In addition,in order to enhance the smooth characteristics of edge details,the robustness and denoising effect of the algorithm are further improved by constructing a space-spectrum global full variational global constraint.3.A non-local tensor low-rank denoising algorithm based on subspace is proposed.High-dimensional hyperspectral images have low-rank characteristics in subspaces.For the subspace feature maps of hyperspectral images,spatial non-local self-similarity and tensor Tucker decomposition are used to perform joint low-rank constraints on the subspace feature maps.It further improves the noise removal ability of mixed noise and water mist noise.