Superresolution Of Hyperspectral Images Based On Spatial-Spectral Information Coordination

With the development of the techniques of sensors, the spectral resolution of remote sensing images is increasing continually. The creation of hyperspectral images is an important leap in the field of remote sensing. Hyperspectral remote techniques keep ahead for the present and a period of time in the future in the field of remote sensing, so the research has been regarded highly at home and abroad. With the increase of the spectral resolution, hyperspectral images can settle many problems that multispectral images cannot do, but its spatial resolution still cannot satisfy the demands. There are a large number of mixed pixels existing in hyperspectral images, which affects quantified analysis and application in other fields. Therefore, it has significant value to increase the spatial resolution of hyperspectral images using super-resolution techniques.In this thesis, a super-resolution algorithm for hyperspectral images based on coordination of spatial and spectrum information is studied. Linear unmixing technique for hyperspectral image is studied firstly. Unmixed result is fed to a super-resolution system. Local analysis of the methods used by edge detection of the region, the mixed-pixel on the edge of the region to improve the use of cellular automaton model for sub-pixel positioning, smooth region is a pixel copy. Then, to achieve the original image of the Hyperspectral the super-resolution processing, the result of positioning use as a reference map for POCS super-resolution algorithm.In this thesis, two high-spectral data are used to verify the effectiveness of the approach. First of all, the linear unmixing technique, sub-pixel positioning, and POCS algorithm simulation ensure that all key technical feasibility. And then, the super-resolution is build for hyperspectral image entire system. Finally using the effective application of the test method, the results of the super-resolution applied to the detection of targets. Certified, super-resolution treated the same false alarm probability of detecting a target probability of a significant improvement. After the super-resolution of hyperspectral image processing, spatial resolution by a large degree of increase, more detailed image of the rich and is conducive to the further image processing, analysis and application.