Research On Synthetic Aperture Radar Image Registration Technology Based On Features

Synthetic aperture radar(SAR)is an active microwave sensor widely used in civilian and military applications because of its high penetration capability,and the ability to acquire data in all-day and all-weather conditions.SAR image registration as a key technology in SAR image processing.The accuracy of SAR image registration directly affects the quality of subsequent SAR image processing,such as SAR image change detection,SAR image fusion,target detection,target tracking and motion analysis.Therefore,it is of great significance to study SAR registration technology in depth.In recent years,feature-based SAR image registration has become the mainstream research direction and research hotspot in the field of SAR image registration because of its rich and diverse features and efficient registration performance.The research content of this paper is mainly focused on Feature-based SAR image registration methods.Although the feature-based SAR image registration method has good registration performance,it still has the following problems:(1)Feature extraction is difficult and the extracted features are unstable and inadequate,which results in the low accuracy of SAR image registration.(2)Feature matching difficulty.After extracting features from the reference image and the sensing image,how to match the features is also one of the difficulties in SAR image registration.(3)How to improve registration accuracy.In feature-based SAR image registration,no matter which feature is selected,the extracted feature will have certain error and instability due to the influence of speckle noise and the limitation of feature extraction method itself,which makes the final registration accuracy not high.Aiming at the problem that feature extraction is difficult and the stability of feature points is insufficient,which lead to low accuracy of registration,this paper proposes a multi-sensor SAR image registration method based on stable convex corner points.This method starts from the region feature,a new region extraction algorithm,Local Stable Extremal Region(LSER)is proposed.It uses a certain proportion of the current regional mean to drive the process of the algorithm can effectively detect the extreme regions in SAR images,the absorption operation obtains more stable extremum regions by ellipse fitting of the extracted extremum regions and merging the adjacent extremum regions.The stable convex corners proposed in this paper are extracted from the strong scattering regions that have been extracted by LSER.These strong scattering regions are fitted to the smallest convex polygons by the Graham scan algorithm,and the stable convex polygons are normalized to the same circular region.Then,the stable convex corner points are detected from these normalized stable convex polygons.Traditional SAR image registration methods extract single feature,the number of features provided by a single feature is often not enough to achieve accurate registration.In feature point matching,the traditional matching method based on local feature description has a low degree of matching,which will produce a large number of mismatched points and mismatched points.In order to solve the above problems,this paper proposes a SAR image registration method based on hybrid feature detection and multi-feature constraint matching.To obtain abundant and stable features,a hybrid feature detection method was proposed involving stable convex corner points and stable local extremum points.As the hybrid features were detected from normalized regions,the method is stable and robust against errors.Multi-feature constraint matching uses different feature constraints to rapidly eliminate mismatches.The features of SAR images are matched with multi-feature constraints from coarse to fine by region constraints,shape constraints,network constraints and spatial relations constraints at different stages,so as to achieve accurate registration.Aiming at the difficulty of SAR image precise registration problem and the unstable feature extraction results in low registration accuracy problem,this paper proposes a SAR image registration method based on particle filter guided by feature.SAR image registration is carried out in three stages from coarse to precise.In the initial registration stage,the region feature and its spatial relationship feature are used to obtain a rough transformation matrix.Thus,the state space can be set in a small range during the initialized particle filter in the coarse registration stage,which greatly reduces the computational complexity and avoids the failure of registration or premature falling into the local optimum.In the rough registration stage,the mismatched particles are eliminated by the mutually exclusive features of strong and weak scattering regions,and the transformation matrix of particle filter is more precise by the guidance of matched feature points,so that the optimal transformation parameters can be searched in a smaller range in the next stage.In the stage of precise registration,the mutual information is used as the measurement value to drive the particle filter algorithm,and the simulated annealing method is used to make the particle filter achieve the global optimization faster,so that the SAR image registration results with higher accuracy can be obtained.