Hyperspectral And Multispectral Image Fusion Based On Decomposition

Because of the rich spectral information,remote sensing hyperspectral images are widely used in precision agriculture,military target recognition,ground object survey and other fields.However,due to the limitation of physical conditions of imaging sensors,the spatial resolution of remote sensing hyperspectral images is relatively low,which is slightly insufficient in specific applications.The fusion of high-resolution multispectral images of the same scene is the main way to solve this problem.Degradation from high spatial resolution to low spatial resolution can be regarded as the process of mixing pixels,while image fusion is its inverse process,which can be regarded as the process of unmixing.Therefore,this paper proposes a hyperspectral image fusion algorithm based on decomposition.In this paper,two modes of matrix decomposition and tensor decomposition are respectively adopted to achieve hyperspectral image super-resolution.The main work of this paper is as follows:The method of image fusion based on image unmixing is widely developed because of the intuitive interpretation of the process of image fusion.This article is based on the fusion idea of unmixing and the content of the research is mainly from the following aspects:First,for the problem of spectral distortion caused by the spatial degradation process of hyperspectral images,the generalized regression neural network model is used to learn the nonlinear mapping relationship between high and low spatial resolution hyperspectral images in the spectral domain.The spectral features of low-resolution hyperspectral images are mapped into a high-resolution space to provide more accurate spectral features for fusion.Secondly,for the problem of the dependence on matrix decomposition in the algorithm based on unmixing,and the instability and initial value sensitivity caused by the non-convexity of the objective function,the spectral features of the image are extracted by clustering instead of unmixing.Improve the stability of problem solving,while effectively improving the efficiency of feature extraction and reducing complexity.Third,for the problem of matrix decomposition focusing on the spectral characteristics and ignoring the spatial relationship,a fusion algorithm based on tensor decomposition is adopted,and the core tensor is used to fully express the correlation between the spectral characteristics and the spatial characteristics.And especially for the phenomenon of diversity and complexity of spectral characteristics caused by the proximity effect in the edge area,the quality of pixel reconstruction at the edge of the image is improved.Finally,experiments were conducted on the Salinas dataset,Cuprite dataset,Indian Pines dataset,and Pavia Center dataset.Comparison with a large number of algorithms proves the effectiveness of the algorithm in this paper.