Signal Processing for Synthetic Aperture Sonar Image Enhancement
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
This thesis contains a description of SAS processing algorithms, offering improvements in Fourier-based reconstruction, motion-compensation, and autofocus. Fourier-based image reconstruction is reviewed and improvements shown as the result of improved system modelling. A number of new algorithms based on the wavenumber algorithm for correcting second order effects are proposed. In addition, a new framework for describing multiple-receiver reconstruction in terms of the bistatic geometry is presented and is a useful aid to understanding. Motion-compensation techniques for allowing Fourier-based reconstruction in widebeam geometries suffering large-motion errors are discussed. A motion-compensation algorithm exploiting multiple receiver geometries is suggested and shown to provide substantial improvement in image quality. New motion compensation techniques for yaw correction using the wavenumber algorithm are discussed. A common framework for describing phase estimation is presented and techniques from a number of fields are reviewed within this framework. In addition a new proof is provided outlining the relationship between eigenvector-based autofocus phase estimation kernels and the phase-closure techniques used astronomical imaging. Micronavigation techniques are reviewed and extensions to the shear average single-receiver micronavigation technique result in a 3 - 4 fold performance improvement when operating on high-contrast images. The stripmap phase gradient autofocus (SPGA) algorithm is developed and extends spotlight SAR PGA to the wide-beam, wide-band stripmap geometries common in SAS imaging. SPGA supersedes traditional PGA-based stripmap autofocus algorithms such as mPGA and PCA - the relationships between SPGA and these algorithms is discussed. SPGA's operation is verified on simulated and field-collected data where it provides significant image improvement. SPGA with phase-curvature based estimation is shown and found to perform poorly compared with phase-gradient techniques. The operation of SPGA on data collected from Sydney Harbour is shown with SPGA able to improve resolution to near the diffraction-limit. Additional analysis of practical stripmap autofocus operation in presence of undersampling and space-invariant blurring is presented with significant comment regarding the difficulties inherent in autofocusing field-collected data. Field-collected data from trials in Sydney Harbour is presented along with associated autofocus results from a number of algorithms.