Adaptive Remote Sensing Image Fusion Based On Variational Method

Remote-sensing images have been widely used in military and civilian fields.With more and more remote sensing information collected by satellites,remote sensing image fusion methods that can make full use of these information and remove redundant information from them become more and more important.These methods are to fuse the panchronic image and multi-spectral image collected by different sensors of the same satellite into a single image with rich spectral and spatial information,so that the observation and processing of the image can be performed.Fused images have been mainly used for ground object recognition,ground feature analysis,urban heat island and so on.Remote-sensing image fusion methods have achieved outstanding results in recent decades,and are mainly divided into three categories.Each of these three methods has its own characteristics and deficiencies.The fusion method based on component substitution is simple and feasible;the spatial information of the fusion result is more than enough,but the spectral information is insufficient.Although the fusion method based on multi-resolution analysis has sufficient protection for spectral information,it is easy to lose more space information in decomposition and reconstruction.So,the fusion image is easy to blur.Fusion method based on variational method has better results in spectral protection and spatial information extraction,but the time complexity is very high,and the noise immunity is insufficient,and it is easy to be interfered by noise.This paper aims at the problems existing in the current fusion algorithm based on variational method.The main research contents are as follows:1.In this thesis,an energy functional term for the assumption of mean gradient difference has been proposed,which is used to improve the fusion method based on dynamic gradient sparsity.In addition,the average gradient difference ratio is calculated during the energy functional solution.It is used to detect the gradient information during iterative solution,and it is determined whether the result has been fitting.We performed corresponding experiments on different satellite data for this method,and the experimental results show that this method can obtain good results in the evaluation of spectral information and spatial information,and the iterative solution process is adaptively terminated,reducing the computation.2.In this thesis,a fusion method based on intrinsic tensors sparse regularization has been proposed,which is used to improve the noise immunity of variational methods.The main application of multispectral image's properties,which are global correlation along spectrum and nonlocal self-similarity across space.the energy function of spatial information extraction is added to the corresponding denoising optimization energy functional,so that the spatial information can be obtained simultaneously in the process of optimization and fusion result can be gained.The experimental results show that this method can achieve better results for images without noise.In the image fusion experiment with noise,the noise immunity is the best among the contrast methods.