高分辨SAR稀疏目标成像研究

发布时间：2018-06-02 12:10

本文选题：合成孔径雷达 + 压缩感知　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：高分辨SAR成像一直是研究的重点问题,这几年兴起的高分辨SAR稀疏成像更是备受关注,这里的稀疏主要是指成像场景中包含少量且很强的散射点,这就是我们后面要讨论的高分辨SAR稀疏目标成像。其目的主要是最大程度的减少成像场景的背景杂波、噪声和旁瓣对目标的干扰,重点关注稀疏目标的成像质量,以达到突出目标,削弱干扰的效果,降低后续的目标检测和识别的难度。基于Nyquist采样定理的传统回波数据采样方法通常获得全采样的数据,导致高分辨SAR的采样率过高,数据量剧增,给数据的存储、传输和实时处理带来了很大的困难。压缩感知(Compressed Sensing,CS)理论的出现,为降低雷达数据采样率,减小雷达平台硬件端的压力,改善雷达成像质量开辟了新的思路。国内外学者将CS理论与SAR成像理论相结合,在基于距离向或方位向的一维CS成像和二维CS成像应用方面取得了一批研究成果,使稀疏目标场景的成像质量有了显著的改善。但是,CS与SAR成像的结合还有很多细节有待进一步深入研究。不同于压缩感知,低秩矩阵重建理论根据成像场景的低秩特性,从矩阵的秩的角度对受噪声干扰和缺损的数据进行恢复。本文对这两种方法在高分辨合成孔径雷达稀疏目标成像中的应用进行了系统的研究,所取得的研究成果为:1.对CS与高分辨率SAR稀疏目标成像的结合进行了研究。针对CS成像中存在的观测矩阵耗费存储大,重建结果时间长以及参数设置复杂的问题,提出了改进的二维CS的SAR成像模型。首先对原始回波数据进行距离徙动(RCM)校正,消除距离向和方位向的二维耦合,然后对观测场景的距离向和方位向分别建立观测矩阵进行观测,这样可以显著减少观测矩阵的存储量和原始数据量,最后,利用改进的迭代硬阈值(IHT)算法对成像场景进行重构。我们利用成像场景服从特定分布的先验知识,简化了阈值参数的求取方法,在成像场景稀疏度未知的情况下,仍然可以获得比较好的二维高分辨稀疏目标成像结果,仿真和实测数据的成像结果都验证了本文方法的有效性。2.对低秩矩阵重建在SAR高分辨稀疏目标成像中的应用进行了研究。不同于CS的向量化处理,低秩矩阵重建理论包括矩阵填充(MC)和低秩矩阵恢复或鲁棒主成份分析(RPCA)理论依据矩阵秩的特性,直接针对二维信号矩阵进行处理,我们将传统的成像算法与低秩矩阵重建理论相结合,提出了一种新的成像框架。首先,我们证明了距离徙动校正(RCMC)后的原始回波数据矩阵的秩与观测场景的秩相等。由此可知,对于低秩的成像场景,经过RCMC的回波数据也具有低秩特征,这是MC在SAR数据中应用的首要条件。接下来,首先进行距离徙动校正,再利用矩阵填充方法对缺损和受噪声污染的回波数据进行补全和去噪,恢复回波数据的低秩特性,并将回波数据分解为两个低维的低秩矩阵,达到降维压缩的目的。最后,根据低秩矩阵恢复理论(RPCA)建立新的SAR成像模型,利用加速近似梯度算法(APG)进行重建,重建结果分为低秩和稀疏的两部分,其中稀疏分量就是稀疏目标的聚焦结果。点目标和实测数据实验都验证了该成像框架的可行性和有效性,并说明了矩阵秩的信息在回波数据处理中有丰富的应用,若能进一步充分挖掘和利用回波矩阵和观测场景矩阵秩的信息,则可以进一步减小原始数据压缩和成像对观测场景低秩特性的依赖。
[Abstract]:High resolution SAR imaging has always been the focus of research. The high resolution SAR sparse imaging, which has been developed in recent years, is more and more concerned. The sparsity of the high resolution is mainly refers to the small and very strong scattering points in the imaging scene. This is the high resolution SAR sparse target imaging which we have to discuss later. The purpose is to minimize the imaging field to the maximum extent. The background clutter, noise and sidelobe interference to the target, focus on the imaging quality of the sparse target, in order to achieve the target, weaken the interference effect and reduce the difficulty of the subsequent target detection and recognition. The traditional data sampling method based on the Nyquist sampling theorem often obtains the total sampled data, resulting in the sampling of high resolution SAR. The high rate and the increase of data have brought great difficulties to the storage, transmission and real-time processing of data. The emergence of Compressed Sensing (CS) theory has opened a new idea for reducing the sampling rate of radar data, reducing the pressure of the hardware end of the radar platform, and improving the quality of radar imaging. The scholars at home and abroad and the SAR imaging theory On the other hand, a number of research achievements have been achieved in the application of one-dimensional CS imaging and two-dimensional CS imaging based on distance or azimuth. The imaging quality of the sparse target scene has been significantly improved. However, there are many details to be further studied in the combination of CS and SAR imaging. According to the low rank characteristic of the imaging scene, the noise and the defect data are recovered from the rank of the matrix. The application of these two methods in the sparse target imaging of high resolution synthetic aperture radar is systematically studied in this paper. The research results are as follows: 1. the combination of CS and high resolution SAR sparse target imaging In view of the large storage of observation matrix in CS imaging, long reconstruction time and complex parameter setting, an improved two-dimensional CS SAR imaging model is proposed. First, the distance migration (RCM) correction of the original echo data is used to eliminate the two-dimensional coupling between the distance and square bits, and then the distance of the observation scene. The observation matrix is observed by the direction and direction, which can significantly reduce the storage and the original data of the observation matrix. Finally, the improved iterative hard threshold (IHT) algorithm is used to reconstruct the imaging scene. We use the imaging scene to obey the prior knowledge of the specific distribution, and simplify the method of the threshold parameter estimation. When the scene sparsity is unknown, a good two-dimensional high-resolution sparse target imaging results can still be obtained. The results of the simulation and the measured data all verify the effectiveness of the proposed method.2.. The application of the low rank matrix reconstruction in the SAR high resolution sparse target imaging is studied. Different from the vector processing of CS, the low rank is low. The theory of matrix reconstruction includes matrix filling (MC), low rank matrix recovery or robust principal component analysis (RPCA) theory based on the properties of matrix rank, directly dealing with the two-dimensional signal matrix. We combine the traditional imaging algorithm with the low rank matrix reconstruction theory and propose a new imaging framework. First, we prove the distance migration. The rank of the original echo data matrix after the RCMC is equal to the rank of the observed scene. Thus, for low rank imaging scenes, the echo data of the RCMC also have low rank characteristics. This is the primary condition for the application of MC in the SAR data. The echo data of the acoustic pollution are complementing and denoising, restoring the low rank characteristic of the echo data and decomposing the echo data into two low dimensional low rank matrices to achieve the purpose of reducing dimension compression. Finally, a new SAR imaging model is established based on the low rank matrix recovery theory (RPCA), and the accelerated approximate gradient algorithm (APG) is used for reconstruction and the reconstruction results are divided. For the two part of the low rank and sparsity, the sparse component is the focusing result of the sparse target. The experiment of the point target and the measured data verify the feasibility and effectiveness of the imaging frame, and show that the information of the matrix rank is rich in the application of the echo data processing. If the echo matrix and the observation field can be fully exploited and used in one step The information of the rank of the scene matrix can further reduce the dependence of the original data compression and imaging on the low rank characteristics of the observation scene.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】