Physics-Informed Optimization Methods of Metasurface and Reconfigurable Antenna Inverse Design for Intelligent Sensing and Imaging Systems

Abstract

Advances in metasurface inverse design have the potential to revolutionize intelligent sensing and imaging systems by leveraging computational optimization and machine learning. This thesis presents a unified exploration of physics-informed optimization techniques applied across three distinct works, each addressing a critical aspect of modern engineering challenges. Specifically, we explore the inverse design of meta- surfaces, from the RF domain to the visible range, uniting the fields of wireless com- munication and optical imaging. Chapter 2 introduces a novel approach to the inverse design of GHz reconfigurable antennas using physics-informed graph neural networks, enabling intelligent beam-forming. Chapter 3 delves into the optimization of a multi- layer broadband metalens for dual-functional color-sorting and polarization imaging, demonstrating significant improvements in optical efficiency and functionality. And Chapter 4 transitions to high resolution 3D imaging, presenting a neural single-shot GHz FMCW correlation imaging system that achieves absolute depth reconstruction with high precision. Together, these works illustrate the versatility and impact of physics-informed optimization, uniting computational design and physics priors to push the boundaries of metasurface technologies and beyond.