ECE PhD Thesis Defense: Unay Dorken Gallastegi

ECE PhD Thesis Defense: Unay Dorken Gallastegi

Title: Computational Methods for Scene Inference From Thermal Hyperspectral Measurements

Presenter: Unay Dorken Gallastegi

Advisor: Professor Vivek Goyal

Chair: Professor Emiliano Dall'Anese

Committee: Professor Vivek Goyal, Professor Janusz Konrad, Professor Lei Tian, Martin J. Stevens

Google Scholar Link: https://scholar.google.com/citations?user=mNXmo0QAAAAJ&hl=en

Abstract: Conventional visible-spectrum imaging faces significant limitations in dark environments due to its reliance on external illumination. Thermal imaging addresses this challenge by utilizing ambient thermal radiation naturally emitted by objects, enabling passive sensing both day and night. Extending thermal imaging to a hyperspectral approach, which captures detailed spectral information, significantly enhances scene perception and identification, particularly in challenging conditions.

This thesis develops computational methods for extracting rich scene information from hyperspectral measurements of ambient thermal radiation. We introduce absorption-based ranging, a passive 3D imaging method that exploits wavelength-dependent atmospheric absorption features present in thermal radiance to estimate object distances, temperatures, and emissivity profiles. Using theoretical analysis based on Fisher information and the Cramér–Rao bound, we systematically evaluate ranging performance under various natural conditions, identifying primary challenges such as minimal temperature contrast between objects and surrounding air, as well as biases arising from reflections of downwelling sky radiance.

To address these challenges, we propose inversion algorithms tailored specifically for natural scenes characterized by minimal temperature contrasts between objects and their surroundings. These methods leverage spectral signatures within hyperspectral thermal measurements to robustly estimate distances, temperatures, and emissivity profiles, significantly improving the accuracy even under conditions with subtle thermal differences. Additionally, our approach accounts for complexities introduced by reflected sky radiance, utilizing distinctive ozone absorption features to differentiate reflections from true object emissions, thus enhancing overall accuracy in practical scenarios. Furthermore, we demonstrate the complementary nature of stereo disparity and absorption-based ranging methods through a stereo hyperspectral configuration, revealing their potential integration to enhance passive ranging accuracy across varying distances.