Study On Registration Algorithm Of Multi-source High-resolution Remote Sensing Image

With the rapid launch and application of remote sensing satellites,the rapid simultaneous interpreting and application of multi-source high-resolution remote sensing images acquired by different sensors has become a hot topic in current remote sensing research.Among them,the automatic registration of multi-source,multi-scale and multi-temporal high-resolution images is one of the key technologies,especially for the high-precision automatic registration between meter-level and sub-meter-level images.Compared with the low-resolution remote sensing images,it is more difficult for high-precision automatic registration of high-resolution remote sensing images,especially the registration difficulty between multi-source and multi-scale high-resolution remote sensing images.The reason is that the physical model and spatial resolution of multi-source and multi-scale high-resolution images are quite different.There are fewer effective point pairs with the corresponding points during automatic matching,and high-resolution images are more susceptible to terrain fluctuations and satellite shooting attitudes.Therefore,relying on the special topics of the national key research and development plan,this paper systematically summarizes and summarizes the principles and methods of image registration in the research and analysis of remote sensing image automatic registration methods,based on the two types of image registration methods of grayscale and feature.Focusing on the feature-based image registration method,using experiments to verify the advantages of this method,and systematically researching and verifying a large number of experiments on the key technologies of multi-source high-resolution remote sensing image registration.The following are mainly achieved Progress in:(1)An algorithm for automatic registration of remote sensing images based on grid index is proposed.In the actual registration process,due to the influence of remote sensing image uncertainty factors such as texture similarity and image noise,the feature points detected on the two images are quite different,which leads to the problem that some non-homonymous feature points are also mistakenly as registration point pairs.The paper introduces the idea of grid index to preliminarily filter the results of SIFT algorithm matching.After eliminating a large number of mismatched point pairs,the random sampling consensus(RANSAC)algorithm is used to optimize,which significantly improves the quantity and accuracy of SIFT matching point pairs in remote sensing image registration,and takes into account the differences of different remote sensing image registration scenes when building grid index.(2)A multi-source,multi-scale,high-resolution remote sensing image registration algorithm based on epipolar geometry is proposed.The problem of poor accuracy of automatic registration is caused by geometric distortion between multi-source and multi-scale high-resolution remote sensing images.In this paper,through epipolar geometry and regional network adjustment,the registration point pairs are encrypted and corrected,and a small number of mismatched point pairs in high-precision registration point pairs are eliminated,which further improves the accuracy of multi-source remote sensing image registration,and the accuracy of image registration is obviously better than the traditional registration algorithm.