Hyperspectral Image Super-resolution Reconstruction Based On Superpixel Segmmentation And Couple Nonnegative Matrix Decomposition

Due to the abundance of spectral information,hyperspectral images play an important role in environmental detection,geological exploration and agriculture.However,the existing hyperspectral imaging equipment is unable to achieve high resolution imaging in both spectral resolution and spatial resolution,resulting in low spatial resolution and mixed pixels in the acquired hyperspectral images.This has greatly affected the application of hyperspectral images in the above fields.The spatial resolution of spectral images can be improved by improving the hardware of hyperspectral imaging,but the cost of the analog method increases dramatically.Compared with hyperspectral images,RGB images in the same scene have relatively high spatial resolution,people often improved the super resolution by fusing the hyperspectral image with low spatial resolution and the RGB image with high spatial resolution under the same scene.The main research contents are as follows:(1)Aiming at the problem that the spatial information of RGB image is not fully utilized in the hyperspectral image super resolution algorithm based on spatial spectral sparse representation,a super resolution algorithm based on super-pixel segmentation is proposed to fuse RGB image with hyperspectral image.The algorithm first uses onlinedictionary learning to acquire spectral dictionaries from hyperspectral images with low spatial resolution.Then,by linear transformation of spectral dictionary,the RGB image dictionary in the same scene with hyperspectral image is obtained.After that,the RGB image is segmented into super pixels,and each small image block after segmentation is sparsely coded.Finally,the whole sparse coding matrix is combined with the hyperspectral image spectrum dictionary to reconstruct the image with high spatial resolution.Experimental results show that the algorithm can make full use of the spatial information of RGB images and effectively improve the spatial resolution of hyperspectral images.(2)Aiming at the problem of mixed pixels caused by the low spatial resolution of the images obtained by the hyperspectral imaging system,a super resolution algorithm for fusion of RGB images and hyperspectral images based on the coupling non-negative matrix decomposition is proposed.Firstly,decomposing the hyperspectral image with low spatial resolution and the RGB image with high spatial resolution alternately,obtains the endelement matrix of hyperspectral image data and abundance matrix of RGB,and finally obtains the hyperspectral image with high spatial resolution.The experiment results show that the visual effect of the fused image is improved greatly.