| With the advantages of all-climate observation, high resolution imaging and etc., conventional synthetic aperture radar（SAR） has been widely used in military and civilian fields. However, it has only two-dimensional（2-D） resolution and the defect of shadow-effect, therefore it cannot obtain spatial 3-D images and cannot observe quickly rolling areas. Linear array SAR（LASAR） is an emerging microwave 3-D imaging technology, has the ability of spatial 3-D resolution and provides more information for applications such as target identification. When working under the downward looking model, it overcomes the shadow-effect and could obtain 3-D images of complex scenes such as urban areas. When working under the forward looking model, it could observe the front downward area and provides information for applications such as autonomous navigation. Thus LASAR has broad application prospects and great research value.At present, the theories and methods of LASAR are still at the early stage, there are many places need to supplement and perfect, and LASAR is facing plenty of problems to be solved:（1） Lack of theories and methods of LASAR under the forward looking model, it needs to deduce the ambiguity function and 3-D resolution for the forward looking model.（2） LASAR has to handle a huge amount of data. It is necessary to develop data processing methods with high efficiency and high accuracy.（3） LASAR has a low resolution in the direction along the antenna array and it has become the bottleneck of its practical applications, thus it is necessary to develop relevant resolution enhancement methods. Sparse reconstruction breaks through the signal bandwidth limitation, and provides a new idea to improve the resolution in the antenna array direction, thus it is necessary to research the resolution enhancement models and methods based on sparse reconstruction for LASAR.In order to overcome the above problems, this dissertation has research on the fast and high accuracy 3-D imaging methods and the sparse resolution enhancement methods for LASAR. It has mainly made the following contributions and innovations:（1） Research on the ambiguity function and the 3-D resolution of LASAR for the forward looking model. First, the ambiguity function is deduced for the forward looking model and the spatial 3-D resolution is analyzed based on the ambiguity function. This could explain the feasibility of 3-D imaging ability of the forward looking model in theory, and the downward looking model could be interpreted as a special case of the forward looking model. Second, a novel resolution projection method is proposed to solve the resolution assignment problem. It makes comprehensive consideration to the contributions of the pulse compression and the virtual aperture in along track（AT） direction to the resolutions in the elevation and AT directions. It could explain the resolution projection more practically.（2） Research on the fast and high accuracy 3-D imaging methods for LASAR under the forward looking model. First, a 3-D chirp scaling imaging algorithm is proposed to format 3-D images and to solve the range walk and the geometrical distortion in the forward looking model. Second, a novel fast and high accuracy imaging method is presented based on back projection（BP） and nonuniform fast Fourier transform（NUFFT） interpolation. It uses the NUFFT interpolation instead of sinc interpolation to acquire high accuracy, and the interpolation precision could reach 10-6. Moreover, combined with the compute unified device architecture（CUDA）, it could greatly accelerate the execution efficiency. Thus, this method is called CUDA NUFFT BP. Third, a fast imaging method is proposed based on random antenna phase centers（PCAs）, as a result, it reduces the number of PCAs to format 3-D images without deteriorations of resolutions and peak sidelobe ratio（PSLR）. It could save about 35 times of computation.（3） Research on the sparse models of resolution enhancement and the evaluation parameter of resolution enhancement ability for LASAR. First, a novel 3-D sparse model is proposed based on extracting strong target areas. It could solve the problem that there exists data interferences between different elevation layers. Besides, several criteria are designed to reduce the size of sparse reconstruction problem, as a result, the computation could be reduced. Second, a fractal dimensional sparse model is presented based on the antenna array dimension. It models along the antenna array direction in equi-range slices to greatly reduce the size of sparse problems. Combined with the spatial variation apodization（SVA）, it improves the ability of resolution enhancement of sparse reconstruction. Third, a novel evaluation parameter for resolution enhancement ability is proposed. It could solve the problem that the traditional definition of resolution is no longer suitable for sparse reconstruction in resolution enhancement. This new parameter based on the error tolerance, the pixel interval and the resolution for comprehensive consideration. It is more in line with actual situations and could provide theoretical guidance to sparse imaging and sparse resolution enhancement.（4） This dissertation proposes the arctangent regularization（ATANR） algorithm for resolution enhancement. It could solve complex-valued sparse reconstruction problems of LASAR resolution enhancement directly, thus overcomes the problem that the 1 regularization solves complex-valued problems by transforming it into real-valued ones, but they are not equal. Compared with the 1 regularization, ATANR is less sensitive to the regularization parameter and has a property of driving its solution to be sparse. In addition, ATANR bases on the idea of approximating 0 quasi-norm, thus it could overcome the poor restricted isometry property（RIP） to some extent in resolution enhancement.（5） This dissertation proposes an iteratively adaptive 1 regularization algorithm for resolution enhancement. It could adaptively adjust the regularization parameter with different observations. Thus it solves the problem that the fixed regularization parameter is not suitable for different observation data. Meanwhile, this algorithm combines the ideas of generalized cross-validation（GCV） and L-curve to select proper regularization parameters, and designs criteria emphasizing on the sparsity to avoid excessive iterations to guarantee sparse solutions. In addition, this algorithm could overcome the problems that GCV could not be suitable for the situation that the number of measurements is smaller than that of scatters, and the L-curve method has a larger computation for it has to examine many regularization parameter values.In one word, this dissertation has researched the ambiguity function and the resolution assignment of LASAR under the forward looking model, has proposed high efficient and high accurate imaging methods for LASAR, has built sparse models of resolution enhancement for LASAR, and has presented the ATANR and the iteratively adaptive 1 regularization methods to handle complex-valued sparse reconstruction problems directly. The research results could provide important theoretical guidance and technical support for the fast and high accuracy imaging and the sparse resolution enhancement for LASAR.
|
PDF Full Text Request