Thesis Defence: Automating Superbubble 3D Segmentation in a Multiphase Interstellar Medium Using Computer Vision

Jing-Wen Chen, supervised by Dr. Alex Hill and Dr. Mohamed Shehata, will defend their thesis titled “Automating Superbubble 3D Segmentation in a Multiphase Interstellar Medium Using Computer Vision” in partial fulfillment of the requirements for the degree of Master of Science in Computer Science.

An abstract for Jing-Wen Chen’s thesis is included below.

Defences are open to all members of the campus community as well as the general public. Registration is not required for in-person defences.

---

Abstract

The interstellar medium (ISM) is a complex and dynamic environment influenced by various astrophysical processes. Among these, superbubbles, the vast cavities filled with hot, low-density gas resulting from multiple supernovae and stellar winds, play a key role in shaping the ISM’s structure and dynamics.

In our study, our objective was to comprehensively analyze the 3D morphology and temporal evolution of superbubbles using computer vision techniques. Initially, we employed basic image processing methods, such as thresholding, to identify low-density regions within density maps corresponding to superbubbles. However, these approaches lacked robustness and failed to accurately represent the intricate 3D structures of the superbubbles.

Subsequently, we explored video object segmentation techniques, leveraging memory modules to capture the continuity of superbubble structures across sequential 2D slices. Despite these efforts, the sequential nature of the input led to a deficiency in the model’s ability to fully grasp the 3D morphology essential for precise characterization.

To overcome these challenges, we proposed a physics-guided U-Net transformer model that effectively captures the complex 3D morphologies and dynamic evolutions of astrophysical structures. Focusing on a specific superbubble, our model successfully generated detailed 3D segmentation masks, enabling visualization and analysis of the bubble’s structural evolution over time. The results provide insights into the superbubble’s growth patterns, energy retention, and interactions with surrounding interstellar matter.

Our findings highlight the potential for integrating state-of-the-art computer vision models into astrophysical research. This interdisciplinary approach not only enhances our understanding of superbubble dynamics but also offers a robust framework for investigating other complex phenomena within the cosmos.