Sparse SAR Target Imaging In Complex Scene

With the motion of the platform,the radar sends the pulse signal to the target to receive the echo of the target at the same time,according to the echo signal to calculate the target complex scattering coefficient to complete the imaging,the imaging process will not be affected by bad weather,with all-weather,all day Imaging and many other advantages.As the high resolution radar image can provide more target information,improve the radar resolution has been one of the main directions of SAR technology development.In order to obtain high resolution radar images under limited computational and storage resources,sparse imaging based on Compressed Sensing(CS)theory has become the hotspot of current research.The theory of compression perception states that for sparse signals,it is possible to accurately recover high-resolution signals with less observations by solving nonlinear sparse constrained optimization problems.Therefore,the sparse scene under the compression perception imaging technology has been rapid development.However,the actual scene is a complex scene with various features and targets and has unique speckle noise,making it difficult to achieve ideal sparse representation of radar echo signals.First,the SAR receiver echo signal is two orthogonal complex signal,the existing sparse reconstruction algorithm for complex signal recovery is not ideal.At the same time,in the linear model of the existing dictionary representation methods such as DCT orthogonal dictionary on the sparse representation of complex data is also a problem.On the other hand,the linear sparse representation model is simple,but the actual imaging model is often non-linear,linear model is only approximate expression,so in the linear model under a very small number of imaging areas can be sparse description,so that the required reconstruction of the observed value Also more.In addition,synthetic aperture radar imaging coverage of a wide range of large areas,but really interested in the target area may only a small part,if both high-resolution imaging will undoubtedly waste a lot of resources.Research on target-driven imaging model can reduce the complexity of imaging in non-interested areas.In view of the above problems,this paper studies the sparse SAR target imaging based on complex scenes,the specific work is as follows:Firstly,a complex scene sparse SAR imaging algorithm based on dictionary learning and decoupling combined OMP is designed.Synthetic aperture radar received echo signal for the complex data,cannot directly use the existing sparse reconstruction algorithm.Aiming at this problem,a joint OMP sparse reconstruction algorithm is proposed to decouple the complex echo signals.The algorithm solves the problem that the real number reconstruction algorithm achieves the target quality is not high,and has better image quality and higher TBR in point target reconstruction.In addition,an improved KSVD dictionary training method is proposed for complex scene SAR.Experimental results show that the imaging algorithm based on dictionary learning and decoupling combined with OMP can obtain high quality imaging results in complex scenes.Secondly,a complex scene sparse SAR imaging algorithm based on nonlinear model is designed.The linear observation model in the compression perception model only approximates the real observation process.The sparsity of the signal is not fully excavated in the linear model.The sampling rate of the signal is reduced under the linear model,but in some specific background,Such as data storage space is very limited,the platform computing power is not very high circumstances need to further reduce the signal sampling rate to meet specific requirements.Considering that the real signal sampling model is non-linear,there must be some deviation from the linear model,so that the observation process can be expressed more faithfully through the nonlinear model.The imaging method based on nonlinear compression perception is sparse by transforming the original signal into the high-dimensional feature space.In the high-dimensional feature space,the signal can have better sparseness,which can more fully excavate the signal itself Sparseness,more effective to reduce the signal sampling rate,while high-quality reconstruction of the original signal.In this paper,the non-linear compression model is used to study the factors that affect the signal recovery in the nonlinear model.The observation matrix for SAR imaging is designed.When the sampling rate of the signal is reduced to 10%,it can be obtained under the nonlinear frame imaging results.The online kernel dictionary learning method can get a better complete dictionary in real time,sparse representation of the signal better,but because the radar platform is often limited storage space,the need to control the size of the training dictionary at the same time in the original training samples on the basis of continuous A new training sample is added.In this paper,a new sample thinning and pruning strategy are designed to achieve better imaging results.Thirdly,a two-step imaging algorithm for sparse SAR is proposed.With the increase of the resolution of synthetic aperture radar,the requirements of the target area are getting higher and higher,but the synthetic aperture radar area is generally larger,the whole area of high resolution imaging will bring a waste of resources.Therefore,the first step imaging in the region before the low-resolution imaging,and then use the segmentation algorithm to determine the candidate target area,the use of adaptive discriminant dictionary learning method for target recognition,accurate positioning of the target;the second step on the target area Resolution imaging.The algorithm not only saves the sampling resources effectively,but also enhances the target information,which is beneficial to the subsequent target processing...