Synthetic Aperture Radar Image Formation For Distributed Target With Sparse Representation

Synthetic Aperture Radar(SAR) is a high-resolution microwave imaging radar and it’s also the most important platform for microwave remote sensing. Not only it has the advantages of all-weather and all-time observation, but also has wide microwave frequency band. The target can be observed in multi-mode, multi-band and multipolarization, thus we can obtain a wealth of geographic information. The traditional SAR imaging algorithm is already quite mature and it has been widely applied in the SAR system. However, the imaging algorithm based on Fourier transform is very sensitive to the phase error, and it is impossible to describe the anisotropic scattering characteristics. To obtain high-quality SAR images, motion compensation of high computational complexity is inevitable, and the system phase error is also difficult to accurately eliminate. A hotspot of the current SAR imaging and feature extraction is to combine the compressed sensing theory and SAR observation model, from which we can obtain the SAR imaging based on sparse representation. Based on sparse representation method, distributed SAR imaging targets were studied in this paper. The main contents are as follows:Firstly, analyze the compressive sensing theory and SAR imaging model and obtain the SAR imaging model based on sparse representation. We adopt the typical scattering components based on scattering center model to study the distributed target imaging, and get the over-complete dictionary by sampling the parametric space of typical scattering components.Secondly, the sparse representation SAR imaging algorithm based on regularization constraints is studied in this paper. We introduce regularization constraints in the cost function of sparse representation SAR imaging problem, and obtain the SAR image under sparse constraints by solving the cost function of quasi-Newton iterative algorithm. Investigate how the regularization constraint operator and ?p norm effect on image point feature and edge feature. Against the problem of parameter space oversampling resulting in excessive dictionary atoms and excessive cross-correlation, we adopt the sequence clustering algorithm to perform preprocessing for over-complete dictionary. It can effectively reduce the cross-correlation of dictionary atoms and reduce the computational complexity.Finally, analyze SAR imaging phase error problem. In practical applications, we often encounter the problem of the phase errors caused by SAR platform jitter and the inaccurate distance estimation between observing scene and the platform owing to track deviation. According to this, we derive the model and obtain the SAR observation model under phase errors. The common phase error is classified into three categories, namely, the one-dimensional phase error, two-dimensional non-related phase error and twodimensional associated phase error. We performed the error correction for observation echoes using sparse drive autofocus algorithm, and under the influence of three types of phase error, we verified the effect of the SAR imaging algorithms and elimination of phase noise based on autofocus sparse driven.