Hyperspectral And Multispectral Image Fusion Based On Dictionary Learning

With the development of remote sensing technology,remote sensing data fusion is widely concerned.Among them,hyperspectral and multispectral image fusion is a research hotspot.The purpose of hyperspectral and multispectral image fusion is to obtain a high spatial resolution hyperspectral image.It solves the problem of low spatial resolution of hyperspectral images obtained directly from sensors.The task of fusion is to maintain the spectral information and enhance the spatial information of the image.Based on this,this dissertation studies the fusion algorithm through the dictionary learning method.On the basis of the existing excellent algorithms,this dissertation proposes two fusion algorithms based on dictionary learning by analyzing the common problems existing in the fusion results.The main work of this dissertation is as follows:(1)In the framework of hyperspectral and multispectral image fusion method based on spectral dictionary,this dissertation proposes a hierarchical dictionary learning algorithm.Because of the complexity of spectral features in the edge detail region of remote sensing image.By learning the dictionary in detail layer and image layer,and combining the two dictionaries to obtain the spectral dictionary,the spectral distortion is minimized in the edge detail region of the image.In the estimation of the coefficients,the detail perception error term and the edge direction adaptive total variation constraint term are proposed.The detail sensing error keeps the spatial and spectral information of the reconstructed image,and the edge direction adaptive total variation is used to calculate the minimum edge direction gradient of the image,so as to achieve the accurate representation of the edge details.(2)The image fusion method based on spectral dictionary will lose spatial information.This dissertation proposes a spatial error estimation algorithm based on sparse representation to solve this problem.According to the preliminary fusion results based on spectral dictionary and Imaging model.The loss of low spatial resolution hyperspectral image and high spatial resolution multispectral image can be calculated,and the spatial dictionary of these two losses can be established to estimate the loss of the preliminary fusion results.Based on sparse representation and spatial error estimation algorithm,the fusion loss is constructed by taking advantage of the particularity of hyperspectral and multispectral image fusion.This fusion framework can be grafted onto any spectral dictionary-based fusion method to make up for the loss of spatial information in the spectral dictionary-based fusion method to the greatest extent.In this dissertation,experiments are carried out on Pavia University and Indian Pine datasets,and the results are compared with the existing excellent algorithms.The results show that the proposed algorithm is very effective in spectral and spatial preservation.