Reconstructia imaginilor SAR de inalta rezolutie / Sparse reconstruction for high-resolution inverse SAR imaging

Synthetic aperture radar (SAR) and inverse SAR play key roles in remote sensing. The Doppler effect, owing to the movement of the radar or target, is key to imaging. This thesis focuses on the signal processing and spectral analysis of SAR data. The Doppler centroid (fDC) is significant for ocean scientific pursuits. Three fDC methods were implemented to address their capabilities in the context of the SAR scene analogy. The de-ramping of TOPS SAR data is designed in the SNAP toolbox, which is a necessary step before the fDC estimation. In addition, the competency of the complex-valued convolution autoencoder is assessed to preserve the original complex SAR properties, including the fDC parameter for ocean surface current analysis. A novel coastline extraction method is proposed using fDC images based on the constant false alarm rate (CFAR) approach. The impact of natural hazards on coastline degradation was investigated using Doppler parameters. The proposed technique was compared with polarimetric correlations and in-situ data. Subsequently, subaperture (SA) decomposition was deployed for ship detection using CFAR. SAs were used to calculate the ship velocity. Expanding on CFAR, the ice cover was delineated using a dual-pol descriptor, and the ice cover retreat velocity was calculated using Euclidean displacement. ISAR provides images of rotating objects. Experiments evaluated the performance of the imaging methods to achieve high resolution. The point spread function, which is an impulse response, demonstrates the performance of each method. Following ISAR imaging, the dual-pol decomposition technique extracted the scattering properties from ISAR images. Eigen-descriptors were utilized to identify the scattering mechanisms through the Lee and Pottier H-alpha plane. This study represents the first case of ISAR images being used for Radar Vegetation Index analysis. Finally, coarse-to-fine estimation, a 2D compressive sensing algorithm, was developed to discard a certain number of columns from the sensing matrix as the reconstruction progressed from coarser to finer scales, thereby achieving high-resolution imaging. This brings us closer to real-time applications for target imaging without any pre-processing or pipeline motion compensation.