Hyperspectral image denoising method based on subspace 4D tensor multimode cascade decomposition

The invention discloses a hyperspectral image denoising method based on subspace 4D tensor multimode cascade decomposition, and the method comprises the steps: projecting an original hyperspectral image into a low-dimensional subspace, representing the original hyperspectral image as a mode-3 product of a coefficient tensor and a basis matrix, and rewriting an observation model; differential continuous regularization constraints are added to the orthogonal spectrum bases of the subspaces; a plurality of non-local similar tensors are extracted and stacked from the coefficient tensor, and denoising of a 4D non-local similar tensor group Zi is completed through a brand new 4D tensor multimode cascade decomposition technology. According to the method, potential low-rank performance and sparsity of the 4D non-local similar tensor group are fully mined in a cascade mode, and mixed noise in a hyperspectral image is effectively removed. 本发明公开了一种基于子空间4D张量多模级联分解的高光谱图像去噪方法，包括：通过将原始的高光谱图像投影至低维子空间中并将其表征为系数张量与基矩阵的mode-3乘积，