Abstract
Synthetic Aperture Sonar (SAS) imaging as an important tool in underwater exploration often has problems of low-resolution quality. The images have blurring effects that degrade sharpness and critical targets could be missed. Resolution enhancement in SAS images has been dependent on traditional methods like brute force, optimization, polynomial, and phase correction autofocus techniques. The high dynamics of underwater environments causes significant temporal and spatial instability in SAS signals, in which various physical (e.g., salinity and water pressure) factors complicate modeling and calibration. Learning-based methods are less often used to enhance image resolution because ground-truth data are challenging to obtain in highly dynamic underwater environments. To improve SAS image quality, the thesis introduces two innovative methods to enhance SAS image resolution without needing external high-resolution ground-truth images. The first method is an ensemble polynomial approach, and the second one is the ridge regularization. Experimental results show that the polynomial ensemble method outperforms existing techniques in suppressing blurring effects. Sharpness of SAS images is enhanced. On the other hand, the ridge regularization method does not improve the image quality as much as that of the ensemble polynomial method.