Deep Learning and Phase Retrieval for Chemical Holographic Imaging: System Inverse Modeling and Sample Property Prediction

With the recent availability of mid-infrared coherent light sources and advances in larger MCT focal plane array (FPA) detectors, discrete infrared imaging (DFIR) enables high-resolution and high-speed infrared (IR) imaging for applications across multiple fields such as clinical diagnosis, forensics, and material science. The problems existing in currently available DFIR imaging systems are examined. Specifically, the fringing artifacts induced by coherent light sources, such as synchrotron radiation and quantum cascade lasers (QCLs). The rigorous mathematical models are given, for the simulation of coherent light interacting with a single interface or multiple interfaces, and materials with different refractive indices. A hybrid imaging system is proposed to address the fringing artifact problem in DFIR when coupled with QCLs. The proposed system is not only able to provide comparable spectral information with high resolution in the spectral domain but also surpasses commercialized DFIR imaging systems and Fourier Transform Infrared (FTIR) imaging systems with superior spatial resolution and faster imaging speed. A more sophisticated chemical holographic imaging system (CHIS) is then proposed based on interferometry and holography. CHIS provides additional information with phase-sensitive measurements based on the DFIR imaging technique. Different forward modeling strategies are discussed, such as forward models for layered homogeneous samples and spherical heterogeneous samples. Potential inverse models for CHIS are also explored using iterative optimization methods to solve for the inverse layered-model problem. For the inverse Mie-Scattering problem, Deep Learning models are applied to seek the optimal solution after spotting the ill-posed nature of the inverse Mie problem. Specifically, an artificial neural network (ANN) is used to predict the sample properties from the reduced form (1-D representations) of the measurement. Evaluations are provided to demonstrate the merit of phase retrieval provided by CHIS through the significantly improved prediction accuracy. The hardware implementation of CHIS, as well as different modules of the implementation, are individually illustrated in detail. The reconstruction process of CHIS is verified using a forward model to demonstrate CHIS's capability of capturing the phase information from the holograms. Finally, preliminary experiments are provided, along with reconstructed data, to demonstrate the additional information -- phase of the field is successfully encoded in the intensity of the hologram, and it can be recovered digitally using analytical solutions.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Electrical Engineering

Grantor

University of Houston

Year dc:date.issued

2020

Author dc:creator

• Ran, Shihao

Advisor dc:contributor.advisor

• Reddy, Rohith K.

Committee members dc:contributor.committeemember

• Mayerich, David
• Shan, Xiaonan
• Prasad, Saurabh
• Larina, Irina V.

• Chemical Holography
• Deep Learning
• Phase Retrieval
• Inverse Modeling

Statement dc:rights

• The author of this work is the copyright owner. UH Libraries and the Texas Digital Library have their permission to store and provide access to this work. UH Libraries has secured permission to reproduce any and all previously published materials contained in the work. Further transmission, reproduction, or presentation of this work is prohibited except with permission of the author(s).

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/10657/7860

OAI identifier oai:identifier

oai:uh-ir.tdl.org:10657/7860