Theory And Method Of Image Location In Support Of Satellite Remote Sensing Data Fusion

In the background of multi-source satellite remote sensing images fusion lacking imaging parameters, the dissertation mainly studies problems on conduction and location of satellite remote sensing images received in non-collaborative manner before image matching. The whole scope covers problems from satellite orbit computation, attitude estimation to image retrieval and location that must be dealt with for geographic location of images, with emphasis laid on fast conduction and location of satellite remote sensing images received in non-collaborative manner and some conclusions are drawn.The work accomplished is listed as follows:1. Definition of time and coordinate systems is presented. Then the relation of gravity potential and gravitational acceleration is deduced, and spherical harmonic function is introduced which results in spherical harmonic series expansion of the Earth's outer gravity potential. The author presents physical explanations of the earth's outer gravity potential based on its series expansion, and obtains the perturbing function that only includes the first degree perturbing potential, which stands for the largest oblateness perturbing effect on satellite orbits.2. Using the operator method for constructing the Adams-Cowell numerical integration formula, the author deduces numerical formulae in term of original functions and also gives formulas of prediction and correction steps.3. The author deduces formulas of the Earth's gravitational acceleration as well as its derivatives with respect to position in fully normalized spherical harmonic coefficients, thus avoiding the gap of quantity level when denoted in non-normalized harmonic coefficients. A detailed scheme for solving satellite motion equations is presented, by which the application of the Adams-Cowell numerical integration method in orbit computation for remote-sensing satellites is demonstrated.4. Two schemes for estimating satellite attitude at the photographing moment are proposed and the feasibility of them is also evaluated.5. The two-dimension moment invariant algorithm is carefully studied, and the invariability of translation and rotation of moment invariant under discrete circumstances is proven, and the effect of proportional factor on the moment invariant is pointed out. The author also finds that although the moment invariant is characterized by the invariability of proportion in case of continuousness, yet that is not the case when discrete.6. After the one-dimension moment invariant of image histogram is studied, the method of image matching is presented which uses the moment invariant of image histogram through matching on a circle template by searching. However, the workload of this method is still too large limited by the moment algorithm itself although dimension-reduction is carried out.7. Base on generalization and analysis of co-occurrence matrix algorithms, a method is proposed that could make fast acquisition of target image texture features and thus implement image retrieval. The method reduces the dimension of co-occurrence matrix and the workload needed for feature extraction. Experiment results show that the efficiency and reliability of image retrieval could be improved prominently. However, the method still demands operations based on grey levels of image, which causes its incapability of fast location when a large scope of image retrieval appears.8. The author clarifies the concept of chain code, and discusses several commonly used chain code description methods. The chain code description of image boundary is obtained through image tracking and thus feature angle points are obtained after angle point detection. Experimental results show that the method is fast, simple, practical reliable and can be applied widely.9. Experimental results are obtained based on analysis and discussion of practical applications of multi-source image fusion.