Research And Implementation Of Registration And Fusion Algorithms For Spaceborne Heterogeneous Images

With the leap-forward development of remote sensing imaging technology,visible light images and SAR imaging have become important means of Earth observation systems.Since the imaging characteristics of the two are highly complementary,they plays an essential function in both military and commercial aspects.In order to effectively integrate the complementary information of the two,it is necessary to align and merge multi-source heterogeneous images of the similar field.However,due to the differences in the imaging mechanisms of the two image sensors,there are many differences in the acquired images.Therefore,how to achieve high precision Stable registration and how to achieve high-quality fusion have become the focus of current research.This paper analyzes the existing heterologous image registration and fusion algorithms,and based on the existing technology,conducts research on SAR and visible light images.The key findings of this paper are:1)To address the issue that the SIFT character extraction operator is unreliable in extracting feature points in the face of large differences in nonlinear radiation between optical and SAR images,in order to obtain stable and repeatable feature points with uniform distribution,this paper proposes a mixed Harris feature selection approach based on nonlinear spread.It constructs a multi-scale space representation based on nonlinear diffusion to better preserve the edge and obtain consistent edge information,utilizes the excellent radiation resistance difference characteristic of phase consistency to calculate the consistent edge structure,and constructs a hybrid Harris feature detection strategy by combining the advantages of both,for detecting stable and repeatable keypoints in scale space.2)Aiming at the problem that the feature descriptors of heterologous images cannot effectively describe the relevant local features,resulting in unreliable feature point matching results,this paper analyzes the characteristics of the feature descriptors constructed based on the gradient domain and the frequency domain,and proposes an improved maximum magnitude index.Graph（MIM）combines GLOH features to describe the spatial structure,and uses the extended phase consistency direction to improve the MIM feature descriptor.On this basis,the spatial geometric structure of the descriptor is expanded to enhance its robustness between image geometric differences;Finally,a single Gaussian robust point matching model is applied to accomplish the spatial geometric relationship modification.The improved method is proven to have better ruggedness and lower time overhead through multiple sets of tests.3)For the problem of target geometric structure information loss and spectral information not prominent in the fusion results of SAR and visible images.In this paper,an image merging approach based on zone partitioning coupled with multi-scale structural character enhancement is presented.The unsupervised region segmentation of SAR images is used to distinguish useful and useless images by dividing SAR images into regions with different radiometric properties;then the source images are decomposed in the NSST domain to characterize different frequencies and directions,and the multi-scale structural features constructed in the multi-scale domain are combined with the margin-preserving softening property of the guided filtering enhance the above high-frequency components,and finally the high and low level fusion regulations are reintegrated.In this paper,five mainstream methods based on multiscale decomposition are compared to verify the effectiveness of the proposed method.4)By evaluating the current remote sensing data processing software systems,the conditions for processing large-scale multi-object type data sets cannot be realized.Therefore,this paper develops related system functions based on the realization of its own algorithm,and designs a set of spaceborne heterogeneous remote sensing images.The handling system has the integrated handling of integrating high-resolution data sets of multiple ground object types,and realizes the integrated design of batch processing,matching,fusion,enhancement,evaluation and export functions of multiple sets of heterogeneous remote sensing images.