SAR Image Registration Of Non-homologous Sea Ice Based On Significant Gray Difference

At present,most researches on sea ice registration at home and abroad use homologous data,but due to the relatively long revisit interval of homologous satellites,the detected features will have limitations.For example,when the selected two sets of data are separated for a long time,some effective and significant features may be lost due to short-time events.This problem can be solved by using non-homologous data,because non-homologous data is more real-time,and data with more features and shorter relative time interval can be more easily obtained.However,registration using non-homologous data also brings additional difficulties.Due to the difference of non-homologous SAR sensors,there are great differences in wave band,polarization mode,imaging mode,etc.,which make the images to be registered have significant gray level difference besides resolution difference.In order to track sea ice drift better,more accurate SAR image registration algorithm is needed.The applicability of feature descriptors and the effect of registration algorithm are two important factors that affect the accuracy of sea ice registration.In this paper,GLDB descriptor is mainly designed and proposed to describe the characteristics of sea ice SAR images with significant gray scale differences,so that it can enhance sensitivity to gray scale differences while ensuring accuracy.At the same time,in the feature matching stage,this paper proposes a two-step rematch strategy based on geometric correspondence matching and gray difference correspondence matching to rematch and verify the matching points.The work of this paper is summarized as follows:Firstly,this paper adopts the method of non-linear scale space construction.Different from indiscriminate filtering of images in linear scale space,it can retain the edge information and important image details of sea ice to the maximum extent,and can better adapt to the problems of inherent noise,resolution and unobvious target significance of SAR images.Secondly,GLDB descriptor for image feature description with significant gray difference is designed and proposed in this paper,which combines the advantages of LDB descriptor with gradient information to enhance the difference and the eight-quadrant circle division extraction method to effectively ensure the extraction accuracy in the main direction.This makes it more suitable for the significant gray image registration studied in this paper.Thirdly,a two-step matching strategy based on geometric correspondence matching and gray difference correspondence matching is proposed in this paper.When error points are eliminated by RANSAC algorithm,the eliminated correct matching points are verified again.This can increase the number of effective matching points,ensure the accuracy of registration,and make the registration result more uniform.Meanwhile,more correct matching points can be obtained for images with significant gray level differences.Finally,in order to verify the advantages of the GLDB-AKAZE algorithm proposed in this paper over the existing algorithms.In this paper,three problems are compared and analyzed:the distribution of matching points is too concentrated,the registration effect of images with significant gray difference is poor,and the number of matching points is overall small.