Title:  Novel Superresolution Methods for Computational Photography and Soil Moisture Remote Sensing

Committee: 

Dr. Romberg, Advisor   

Dr. Lanterman, Chair

Dr. Bras

Abstract:

The objective of the proposed research is to solve two real-world superresolution problems of fundamental importance: extending the resolutions of non-diffraction-limited imaging systems, and estimating soil moisture globally at high spatiotemporal resolution. Towards solving the first problem, we previously demonstrated that augmenting a conventional imaging system with a programmable mask and defocused lens enabled superresolved imaging by factors greater than 4x without the use of mechanical motion or an image model, although this required an inconvenient calibration process. To make this method truly general, we aim to solve the blind superresolution problem that arises when the blur is unknown by leveraging modern image models and requiring that the blurring operator be physically realistic. For the second problem, we aim to combine deep learning for single-image spatial superresolution of soil moisture with numerical modeling to produce physically consistent, high-spatiotemporal estimates of soil moisture at a global scale. In the process we will create an extensible framework for superresolving soil moisture, and address key issues such as the handling of missing pixels in network training and operation.