Empowering Foundation Models for Unsupervised Bi-Temporal Analysis in Satellite Imagery

The School of EECS is hosting the following PhD progress review 1 seminar

Empowering Foundation Models for Unsupervised Bi-Temporal Analysis in Satellite Imagery

Speaker: Yiyun Zhang
Host: Prof Helen Huang

Abstract: Bi-Temporal Remote Sensing (BTRS) analysis aims to compare satellite imagery captured at the same geographical location but at different points in time. Empowered by deep learning techniques, BTRS has demonstrated great potential in various tasks, including change detection and disaster assessment in areas of interest. While deep learning has enhanced BTRS analysis, many state-of-the-art models and methods often rely on numerous labels and annotations, resulting in a huge burden on human labour and time. In this study, we propose novel unsupervised approaches for BTRS analysis, specifically focusing on change and damage detection in buildings, without the need to use any annotations. By leveraging pre-trained foundation models in computer vision and addressing domain gaps in satellite imagery, our research yields promising results for both tasks, showcasing the potential for practical deployment in real-world scenarios.

Bio: Yiyun Zhang is an HDR student at the University of Queensland, under the supervision of Prof. Helen Huang and Dr. Xin Yu. He received a B. InfTech (Hons Class I) at the University of Queensland in 2022. His research interests encompass remote sensing in computer vision and vision-language models.