Sparse Signal Processing Techniques Of Spaceborne SAR Imaging And Moving Target Detection

As an active tool of microwave remote sensing, spaceborne Synthetic Aperture Radar (SAR) system, which has the abilities of wide-area and high-resolution observation in all-day and all-weather condition, has played an important role in military and ciliv activities such as battlefield reconnaissance, terrain mapping, resource exploration, environment survey, and so on. In recent years, with the development of radar technology, the system with single antenna has stepped into the age of multi-dimensionality observation with the systems of multi-antenna, multi-polarization, multi-angle and multi-wave. This improvement represents that the cooperative radar network has been set up with the information obtained from the sensors in land, sea, sky, space and electronics, which would greatly improve the performance of the traditional radar systems and the ability of information sensing. However, in the background of multi-dimensionality observation, the huge amount of data that needs to be storaged and transmitted is required to be considered and solved, and moreover, the efficiency of signal processing methods needs to be further improved. Therefore, in this dissertation, some key problems in SAR imaging and ground moving target indication (GMTI)are addressed. With the support of National Basic Research Program of China (973 Program), some novel technologies using the framework of sparse representation has been researched, where the redundancy between signals are used to solve some practical issues like high-resolution wide-swath imaging and sparse moving targets detection, and these research results would provide support for spaceborne radar observation technology of our country.The main content of this dissertation can be summarized as follows:1. For the spaceboren SAR system, the echo is formulated, where the atmosphere error and slant range error is analyzed and compensated. Firstly, considering that the track of satellite is higher as well as the transmitted frequency, the influence of ionosphere cannot be neglected any more. Therefore, the phase error and the slight deviation of visual angle are detailedly analyzed. By effectively estimating, the error can be compensated and the influence of atmosphere can be alleviated. Furthermore, an accurate calculation method of longitude and latitude is proposed for the region of illumination, which would be more exact to describe the relationship between the antenna and the gound reflective scatters. Besides, to address the high-order item of the instantaneous slant range, the formation in range-frequence and Doppler domain is deduced. At last, the influence of Doppler ambiguity on Keystone transformation is analyzed and a simple method is proposed to estimate the ambiguity number.2. To conquer the contradiction of high-resolution wide-swath (HRWS) in the single antenna system, a compressed sensing (CS) based sparse imaging method is proposed. By using the digital beamforming technology, different sub-swaths are illuminated randomly and the under-sampled data are collected. Considering the different sample series among different sub-swath, the sparse dictionaries are designed separately. At last, the CS based method is used to achieve the HRWS imaging. Compared with the TOPSAR model, the CS based method has a higher resolution. Moreover, the result can be futher used to achieve moving target detection3. A low-rank matrix completion (LRMC) method is proposed to deal with the under-sampled SAR imaging problem. This new theory aims to solve the optimization model of the energy of the matrix, i.e. the sum of singular values, which is similar to the principle of CS. By compensating and reconstructing the range-compression signal to make the structure satisfy the low-rank property, the low-rank minimization problem can be effectively solved. As a result, the unsampled information can be well estimated. Compared with the CS based method, LRMC model is a non-parametric method. Therefore, it does not need to design the sparse dictionary and can obtain an accurate estimation result.4. A robust principal component analysis (RPCA) based method, which combines both the sparsity and the low-rank property, is proposed to detect the moving target in the multi-channel SAR system. In the ideal case, only a stable phase caused by channel distance is remaining between any two channels. After image registration, the stationary clutter would have a high correlation between channels. Based on this property, a new matrix is constructed, which is basically low-rank. However, the moving target would destroy this low-rank property since it has its cross-track velocity that differs from the clutter. Therefore, by using both the sparsity and the low-rank property, the moving target can be effectively detected and extracted from the clutter background. A low-rank clutter matrix and a sparse moving target matrix are separately obtained. Moreover, to have a more practical application, fast detection in range-compression domain and a preprocessing method in range-Doppler domain are separately proposed which addresses the target contamination problem in the covariance matrix estimation. At last, a prior information of shape constraint, which is estimated from the system parameters, is utilized to improve the accuracy of target detection.