Tailored Algorithms for Synthetic Aperture Image Formation and Analysis

Open Access
Author:
Mckay, John David
Graduate Program:
Electrical Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
November 13, 2018
Committee Members:
  • Vishal Monga, Dissertation Advisor
  • Vishal Monga, Committee Chair
  • Jesse Louis Barlow, Committee Member
  • Jeffrey Louis Schiano, Committee Member
  • Necdet S Aybat, Outside Member
Keywords:
  • Synthetic Aperture Sonar
  • Synthetic Aperture Radar
  • Deep Learning
  • Transfer Learning
  • automatic target recognition
Abstract:
Both sonar and radar have seen incredible advances in image quality with the increased usage of synthetic aperture array designs. This clever signal processing technique allows for images from sound or electromagnetic waves with spatial resolution in the centimeters even when capture from hundreds of meters away and over scenes that span several square kilometers. The following thesis looks to investigate how synthetic aperture images are made and how they are used for automatic target recognition problems. We provide three contributions: (1) an improved image formation scheme based on solid mathematical work specifically designed towards the Fourier domain, (2) an automatic target recognition method that leverages sparsely constrained compressive sensing towards noise and training-size-robustness, and (3) work detailing several ways to utilize deep learning for synthetic aperture sonar image classification. Each contribution outlines a potential strategy for both sonar and radar modalities with a bend towards real- world applicability. Indeed, each our or imaging and image classification schemes work towards improvements that tackle actual problems like data-starved imaging, image attenuation, and data imbalance.