| Remote sensing image fusion obtains a more informative and accurate fusion image by complementary integration of multiple remote sensing images,the fusion image also has comprehensive image features and a high utilization rate of the data.This thesis mainly researches the remote sensing image fusion of multispectral(MS)images and panchromatic(PAN)images.By fusion of MS images and PAN images,the spectral information of images can be preserved and the spatial resolution of images can be improved,exerting advantages of various sensors.The MS image and PAN image fusion methods are studied in this thesis,in view of the traditional fusion method about the resolution of MS image abtained by interpolation method to get the matching image is low,the super-resolution reconstruction method is employed to improve the resolution of the MS image,then the reconstruction of the MS image and PAN image separately are fused with Laplacian pyramid fusion and guide filter fusion,The main work of this paper are as follows:The traditional fusion method is discussed,and four fusion methods,such as IHS fusion method,PCA fusion method,GS fusion method and Brovey fusion,are compared.Aiming at the method of image super-resolution reconstruction,based on MS image,three reconstruction methods,including the interpolation method,SRCNN reconstruction method and FSRCNN reconstruction method,are verified and analysed by experiments.In this thesis,two fusion methods of Laplacian pyramid fusion and guide filter fusion are studied.Based on super-resolution reconstruction of MS image and Laplacian pyramid fusion method,CNN feature extraction was carried out on PAN image and the reconstructed MS image to obtain the details of the two images.Laplacian pyramid decomposition was carried out on the two images respectively and the extracted details were fused.Based on super-resolution reconstruction of MS and guided filter fusion method,using guided filter to extract detail information of PAN image and the reconstruction MS image,the extracted detail information is adaptively enhanced and injected into MS image to obtain fusion image,and introduces three filter method,based on the guided filter,based on the fast guided filter,based on the rolling guided filter,the experiment compares three methods,from the point of subjective and objective analysis,the method based on guided filter to get the fusion result is better by contrast of the others.The data from three different remote sensing satellites are selected for fusion experiments,including Geo Eye-1、IKONOS、World View-2.The IHS algorithm,PCA algorithm,GS algorithm,wavelet transform algorithm are selected as contrast methods.Through subjective visual analysis and objective data analysis,the proposed methods of Laplace pyramid fusion and guided filter fusion method have the best effect than the contrast method,and the spectral information of the fused image obtained by guided filter fusion method is the most abundant,and the image detail features are also the most abundant. |
