One of the most advanced and widely used remote sensing imaging systems is the Synthetic Aperture Radar (SAR). SAR radars form long and synthetic apertures by moving their platforms and using a small antenna, and so they make the target images in the form of gray pixels in the surface or volume by extracting the radar cross sectional diagram of the targets. Despite of many advantages of this type of imaging, the large number of samples is one of the main problems of SAR imaging, because it yields to heavy processing to form the final image. Therefore, any innovation and method that leads to a significant reduction in memory size and computational complexity will be valuable. Introducing Compressive Sensing (CS) theory and the possibility of reconstructing sparse signals with a much smaller number of samples than that for the Nyquist rate, we can overcome with the mentioned problems, and so the use of CS in SAR received much attention. In fact, because many of the coefficients in the SAR signal are close to zero in a specific space, we can consider the SAR signal to be sparse or compressible. Also, because the received signal contains strong reflectors, such as man-made devices, and these strong reflectors are placed next to several cells and affect adjacent cells, the SAR signal is more consistent with the definition of block sparse. Therefore, we can use block methods known as Block Compressive Sensing (BCS) to reconstruct SAR images. In this thesis, in addition to reviewing the various compressive sensing methods used in SAR radar, first, a new algorithm is introduced based on the concept of BCS; namely, Block Norm Regularized Orthogonal Matching Pursuit (BNROMP), which is a generalization of the ROMP method considering the concept of block sparsity. In the following, we use the proposed algorithm to form SAR images based on linear frequency modulated waveform and step frequency waveform to eva‎luate the performance of this algorithm in practice and in reality. We will see that using the BNROMP algorithm, images with higher resolution than those for the standard image algorithms and other CS-SAR or BCS-SAR methods can be achieved. In addition, to better eva‎luate the performance of the algorithm, we use real data to see the ability of the algorithm in the real world. In this case, the BNROMP algorithm is much more superior than other algorithms.

محمود مدرس هاشمي

احسان يزديان، مجتبي بهشتي