MSc Thesis Defense: Elman Ghazaei, TOWARDS EFFICIENT AND DOMAIN-GENERALIZED MODELS FOR REMOTE SENSING, Date & Time: June 3, 2026 – 2:00 PM, Place: FENS 2019

TOWARDS EFFICIENT AND DOMAIN-GENERALIZED MODELS

FOR REMOTE SENSING

Elman Ghazaei
Computer Science and Engineering, MSc Thesis, 2026

Thesis Jury

     Prof. Erchan Aptoula (Thesis Advisor)

  Assoc. Prof. Hüseyin Özkan

 Prof. Koray Kayabol

Date & Time: June 3, 2026 – 2:00 PM

Place: FENS 2019

Keywords: Domain Generalization, Semantic Segmentation, Remote Sensing, State Space Models, Deep Learning

Abstract

The rapid expansion of Earth observation technologies has led to the creation of massive remote sensing image archives, increasing the demand for efficient and reliable analysis systems. However, the performance of deep learning models often degrades under domain shifts caused by variations in imaging modalities, geographical regions, and acquisition conditions. To tackle these challenges, first, I propose a multi-level adversarial strategy for semantic segmentation that focuses on extracting domain-invariant features at multiple levels and enforces progressive domain invariance from earlier to later network layers. Second, I introduce a feature disentanglement-based model that progressively separates content and style representations, emphasizing domain-invariant features through a plug-in module. Third, I develop an efficient change detection framework based on state space models, achieving state-of-the-art performance while using only a fraction of the parameters required by existing methods. Finally, I explore DG in the context of Specific Absorption Rate-based visual question answering, where I investigate how incorporating common geographical textual features can improve the generalization capability of Vision Language Models. I conduct extensive experiments across several benchmark datasets. The results show that my proposed techniques substantially enhance the generalization capabilities of the underlying models, providing a strong foundation for developing reliable, domain-agnostic tools for remote sensing image analysis.