Research On Algorithms For Improving Spectral And Spatial Resolution Of Remote Sensing Images

With the characteristics of containing considerable information,remote sensing images can effectively help people observe surface information for feature mapping,land cover classification,and change detection.In recent years,the combination of remote sensing images with sense classification,target positioning,target detection and many technologies has produced valuable applications.Nevertheless,there are still the following problems to be solved: the requirements of remote sensing images are costly;The spectral resolution of obtained panchromatic(PAN)and the spatial resolution of multi-spectral(MS)images are low.In order to solve the above problems,domestic and foreign scholars have carried out related research.Abundant pan-sharpening and super-resolution(SR)methods have been proposed.Pan-sharpening requires coupled PAN and MS images,but these images are expensive,and he color transition of the fusion result is unnatural;SR cannot increase the spectral resolution of PAN.So,this thesis would like to design a multi-task neural network to improve spectral and spatial resolution of various remote sensing images for enhancing the readability of unpaired remote sensing images.Image SR algorithms are relatively mature,while image colorization algorithms are scarce.So,this article first explores the feasibility of colorization for remote sensing images;then use the same model to improve the spectral and spatial resolution of remote sensing images.The research content of this article includes the following two aspects: 1.An end-to-end automatic remote sensing image colorization model is proposed to help improve the spectral resolution of remote sensing images;2.There are four different tasks solved by a network structure: enhancement of remote sensing images spectral resolution,enhancement of remote sensing image spatial resolution,enhancement of remote sensing image spectral and spatial resolution,and pan-sharpening.The relevant experiments and method comparisons are carried out.The experiments show that,compared with the comparison algorithms,the image quality obtained by the proposed model is excellent in both subjective and objective evaluation indicators.