Spatiotemporal Remote Sensing Image Super-resolution Reconstruction Based On Multi-scale Detail Enhancement

Spatiotemporal remote sensing image super-resolution is always the processing of several low resolution observed images with complementary information,and super-resolution technique that reconstructs one or more high resolution images.Although the diversity of terrestrial remote sensing image data acquisition is increasing,the spatial resolution of the domestic satellite remote sensing image still has a certain gap with foreign countries due to the constraints of hardware technology.The ZY-3 remote sensing image is selected as the experimental data,and the research is conducted based on the similar and complementary information obtained from ZY-3 image,and a targeted solution is proposed for the spatial resolution enhancement of the ZY-3 satellite remote sensing image.The core technology of spatiotemporal remote sensing image super-resolution is: high precision image registration,detail enhancement of remote sensing image and complementary information extraction from remote sensing data in time domain.Thus,in order to promote image texture detail level and spatial resolution to the largest extent,a novel method is proposed,which is suitable for the satellite remote sensing image with high resolution,high fidelity and more details.The main contents of this dissertation are as follows:First,a tiny facet primitive spatiotemporal sequence remote sensing image registration theory and method is studied.To solve the accuracy problem of remote sensing image registration which is influenced by the uneven distribution of control points,a tiny facet primitive remote sensing image registration algorithm based on the optimized Delaunay Triangulation is proposed in this paper.The Rational Function Model(RFM)and Digital Surface Model(DSM)are used for image ortho rectification.Then,feature points are matched and the edge grid points are multiple constrained,which are used to build the initial Delaunay Triangulation,Finally,the tiny facet primitive image registration is achieved.The experimental results show that the accuracy of image registration can reach sub-pixel level,and effectively solves the problem which exists in remote sensing images with complicated geometric deformation,and it can be used for super resolution reconstruction of remote sensing images.Second,remote sensing images super resolution reconstruction based on multi-scale detail enhancement is proposed.The existing methods such as the lack of texture feature representation and the lack of high frequency details,so that the spatial resolution of the reconstructed image is limited,based on the ZY-3 remote sensing image data acquired by the same time,the remotesensing images super resolution reconstruction based on multi-scale detail enhancement is put forward.The sequence images are decomposed into smooth information which contains coarse-scale image information,and the detail information which contains small-medium-scale information,On the basis of sampling,a texture detail enhancement function is built to improve the scope of small details.Then,the multi-scale information are fused,and a local optimizing model is built to further promote the premier image quality.The experiments show that the entropy index improved about 0.51 bits,and the EME index improved about 4.46,the texture detail of the reconstructed image is improved obviously.Third,super-resolution reconstruction based on the detail enhancement of spatiotemporal remote sensing data is put forward.In order to improve the spatial and temporal complementarity of spatial and temporal data,a new super-resolution reconstruction method based on the detail enhancement of spatiotemporal remote sensing data is proposed.The reference image is determined,and the normalization process of temporal is utilized,the Iterative Back-Projection method was used to reconstruct the original high-resolution images,and the reconstructed images were decomposed.Then the missing high-frequency details information was supplemented in the process of reconstruction,and the edge sharpening was avoided at the same time by constructing the optimized detail enhancement function.Finally,the high-resolution remote sensing image was obtained by conducting weighted fusion based on structural similarity index.The experiments show that the entropy index improved about 0.51 bits higher than that of the traditional reconstruction method,and the EME index is increased by about 1 times.Fourth,image quality assessment method for super-resolution reconstruction is researched.Given that most methods rely on using high resolution image as the reference image in the existing quality evaluation methods of super-resolution reconstruction,the quality evaluation of super-resolution reconstruction without reference image based on remote sensing image is proposed.The structure similarity,the local gradient features similarity,and the reconstructed image detail enhancement index are calculated,remote sensing image super-resolution reconstruction results is evaluated comprehensively,the accuracy and robustness of the evaluation index are verified by experiments.In summary,aiming at the different remote sensing data,different algorithm is used to take the theory of reconstruction,effectively to solve the spatial resolution of remote sensing image promotion question,there is important theoretical significance and practical engineering value for the development of remote sensing image processing theory,upgrading and improving thequality of domestic satellite remote sensing images,to improving the utilization rate of the remote sensing image.