Radar Target Structure Feature Enhancement Based On Regularization Framework

Synthetic Aperture Radar(SAR)can observe and realize high resolution imaging all day and all weather,which provides important technical support for military and civil applications,such as missile precise guidance,air traffic control and meteorological detection of civil aviation.The rapid development of high resolution SAR imaging mainly comes from the appearance of Compression Sensing(CS).CS theory breaks through the limitation of Rayleigh resolution,greatly promotes the accuracy of SAR imaging,and makes it possible to obtain more elaborate/weak scattering structure features of observation targets in SAR imaging.However,the fast developing CS algorithm such as threshold iteration algorithm can not protect the elaborate/weak scattering structure of the target in echo signal recovery,which will lead to SAR losing the advantage of extracting fine features from original high-resolution imaging,and the operation rate of traditional sparse recovery algorithm needs to be improved.In this thesis,the prior information of different kinds of target solutions is used to introduce appropriate regularization term to characterize different SAR scene features,so as to further improve the accuracy of high resolution SAR imaging technology.The specific research contents of this thesis are as follows:1.In order to solve the problem of slow convergence rate of traditional sparse imaging algorithm,Greedy Fast Iterative Shrinkage-thresholding Algorithm(Greedy-FISTA)is proposed to recover SAR moving targets sparsity by combining greedy idea.The method improves the traditional Fast Iterative Shrinkage-thresholding Algorithm(FISTA)by heuristic adaptive restart technology,and determines the optimal step length of each iteration as the step length of soft threshold update and under the constraint of convergence condition,the optimal convergence value is achieved by iterating the complex data of SAR moving targets with large amount of high dimension data.The comparison of convergence speed and simulation and experimental data show that the proposed method has superior performance in convergence speed and SAR moving target sparse recovery.2.In order to solve the problem that only sparse prior is considered in Greedy-FISTA,which is easy to lose the elaborate scatterer structure information,a two-layer sparse group lasso of alternating direction multiplier method(SGL-ADMM)based on Euclidean distance is proposed,the idea of variable splitting in generalized ADMM is used to construct double penalty terms.l1 norm is introduced to represent intra block sparsity,and l1/lF norm to represent inter block sparsity and intra block structure smoothness.The corresponding regularized proximal operators are obtained respectively,and the global optimal solution is obtained by iteration with the idea of "decomposition-coordination" to achieve sparsity and global structure smoothing.The performance of the proposed algorithm is verified by the qualitative imaging comparison with the traditional method and the quantitative comparison with the phase change thermogram.3.The SGL-ADMM based on Euclidean distance block is more "inflexible" in complex morphological scenes,and there is still room for improvement.Therefore,based on ADMM multi task framework and inspired by the ability of mathematical morphology to fit the contour,this thesis proposes a morphological auto-blocking alternating direction method of Multipliers(MAB-ADMM)based on Spatially Variable Morphology(SVM)geodesic distance to adaptively block the echo data.In this method,lM/lF regularization structure penalty term based on morphological block is constructed to represent the target clustering features,and l1 regularized linear regression is used to perform sparse representation.Compared with SGL-ADMM,the proposed algorithm has better structure recovery ability.Unlike MAB-ADMM,which introduces morphology into the construction of block and mixed norm,this thesis proposes a new method of structure guaranteed SAR(SG-SAR)imaging,in which the regularization penalty term is morphological norm.Only morphology is used to construct morphological regularization.In this method,the morphological norm is constructed by SVM operator,and the secondary gradient operator is derived.In order to adapt to complex scenes/targets,SG-SAR adopts the following method This method is similar to MAB-ADMM,and the effect is the same.The experimental results show that the proposed algorithm has good performance in SAR high-resolution imaging.