| Summary: | Synthetic Aperture Radar (SAR) is a dedicated high-resolution sensor with imaging capability in all weather and day-night conditions and has been employed in several earth and interplanetary observation applications. A significant characteristic of this system is the generation of a large amount of data that involves major problems related to on-board data storage and downlink transmission. The near future SAR satellite missions planned would be pushing downlink data bandwidths to prohibitive levels, which dictate efficient on-board compression of raw data. Due to the limitation of the on-board resources in the satellite, it is desirable to have computationally efficient encoder. In this thesis we address the compression of complex-valued SAR raw data in the Compressed Sensing (CS) framework, in which the encoder is simple whereas the decoder is computational expensive. CS is an emerging technique for signal measurement and reconstruction that takes advantage of the fact that many signals are sparse under some basic or frame. The measurement of the signal in the CS framework is obtained by taking a small number of projections of the signal onto an incoherent basis. For the SAR raw data compression here we have considered a simple encoder, with a 2D-FFT followed by a random sampler. The reconstruction of the sparse coefficients of the signal from these projections is then based on the sparsity induced optimization techniques like Orthogonal Matching Pursuit (OMP) and iterative reconstruction methods. We demonstrate empirically that the CS framework for compression of complex-valued SAR raw data is effective for the cases when the SAR image is sparse in the spatial domain. We also address the limitations of this framework while dealing with actual satellite images due to lack of good sparsifying transforms for the complex-valued data. In this thesis, we present a new algorithm based on regularized iterative algorithm for finding sparse solution for the complex-valued data in which the regularization parameter is adaptively computed in each iteration. The effectiveness of the new algorithm is compared with existing methods like Basis Pursuit, OMP, etc with both real and complex data set.
|
|---|
