A Study Of High Resolution SAR Imaging Of Sparse Targets

High resolution SAR imaging has been the focus of research questions, in recent years the rise of high resolution SAR sparse imaging is of concern, where sparse mainly refers to the imaging scene contains a few strong scattering from the target points, which is that we high resolution SAR imaging of sparse targets discussed later. It tries to reduce the interference from background clutter, noise and sidelobe for the targets in a great extent. And our main focuses are the quality of imaging of sparse targets and achieve the effect of similar to target detection and improve the accuracy of the subsequent target detection and recognition. Traditional acquisition method of raw echo data needs full sampling data based on Nyquist sampling theorem. This leads very high sampling rate for high resolution radar and the fast increasing of the volume of raw echo data and a lot of difficulties of storage, transmission and real-time processing of data. The emergence of compressed sensing theory opens up a new thought to reduce the sampling rate of raw echo data and the pressure of the hardware of radar platform and improve the quality of radar imaging. Many scholars at home and abroad combine radar imaging theory with compressed sensing theory. And they has been acquired a series of research achievements such as one-dimensional(1-D) compressed sensing imaging method of azimuth or range direction, two-dimensional(2-D) compressed sensing imaging method etc. These methods can significantly reduce the side lobe and improve the resolution of the targets. However, there are still many details need to be further study. Unlike compressed sensing, low-rank matrix reconstruction theory recoveries the undersampling data based on the low-rank property of observation scene. In this paper, we have studied the application of high resolution SAR imaging of sparse targets with these two methods. And the research results obtained as follows:1. The two-dimensional(2-D) SAR imaging method is studied based on compressed sensing. In view of the problems that observation matrix costs much storage and the reconstruction of results costs time, we proposed the improved 2-D imaging method of SAR based on compressed sensing. First, range cell migration(RCM) is corrected to eliminate the 2-D coupling of azimuth and range direction. Then we set up the observation matrix for azimuth and range direction, respectively. And then we can acquire the 2-D undersampling raw data and significantly reduce the storage ofobservation matrix and the volume of raw data. At last, we reconstruct the observation scene with the improved iterative hard threshold(IHT) algorithm. We use the prior knowledge that the imaging scene follows specific distribution to simplify the calculating method of the threshold parameter. This new method make us still can obtain the similar high resolution two-dimensional sparse target imaging results using traditional iterative hard threshold algorithm under the condition of unknown sparse degree of observation scene. Through the experiments of simulations and real data, the validity and good performance has been proved.2. The SAR imaging method is studied based on low-rank matrix reconstruction. Unlike the vector processing of compressed sensing, low-rank matrix reconstruction includes matrix completion(MC) and robust principal component analysis(RPCA) and directly processes the 2-D signal matrix based on the property of matrix. We combine the traditional imaging method with low-rank matrix reconstruction theory and propose a new framework of SAR imaging. First, we proved that the rank of the raw data after RCMC is equal to the rank of observation scene. Hence, we can know that the raw data after RCMC is also low-rank according to the low-rank property of observation. This supports the condition for the application of MC method in SAR imaging. First, range cell migration(RCM) is corrected to eliminate the 2-D coupling of azimuth and range direction. Then we complete the missing elements and denoise to recover the low-rank property via MC method and divide the raw data into two low-rank matrixes for the purpose of compressing. At last, we set up a new imaging model based on RPCA. Then we use accelerate approximate gradient(APG) algorithm to recover the observation scene. The imaging results are divided into two parts. One part is low-rank and another is sparse. The sparse part is the targets we focus and the low-rank part includes the background clutter and noise. The simulations and real data experiments both proved the validity of the method and state that the information of matrix rank has abundant application in the processing of raw data. If we can further digest the information of matrix rank, the influence of the low-rank property will be reduced greatly for the compression of raw data and imaging.