CRI / Cambridge Centre for Earth Observation Seminar
Zhengpeng (Frank) Feng, Department of Computer Science and Technology, University of Cambridge
Barlow Twins Foundation Model (BTFM) for Earth Observation
Integrating multi-source, multi-temporal satellite EO data into a globally consistent, high-resolution map is a long-standing challenge. Recently approaches use AI to create high-dimensional representation maps, also called ‘foundation models.’ Such a foundation model could transform conservation by helping experts measure habitat loss and gains, spot plant disease, and measure carbon storage. Such a model could also detect illegal logging, assess how well habitats might cope with climate shifts, find the best spots to bring back native species, and gauge if conservation efforts are working—while cutting down on costly field work in hard-to-reach places. Existing foundation models, which process patches or images as inputs, often struggle to capture fine-grained temporal relationships. To address this, we present a self-supervised framework based on Barlow Twins to create a pixel-oriented foundation model that naturally incorporates temporal relationships. Preliminary results demonstrate that our model outperforms other foundation models (including current SOTA model, Google’s Embedding Fields Model) in classification and regression tasks. By bridging multi-modal satellite data into a harmonized latent space, our approach unlocks new opportunities for monitoring planetary-scale processes with higher precision.
Zhengpeng (Frank) Feng is a PhD student in the Department of Computer Science and Technology at University of Cambridge. His research sits at the intersection of computer science and Earth science, with a focus on machine learning applications in remote sensing. Recently, he and team won the Better Future Award at the Computer Lab’s Hall of Fame Awards for their app, GreenLens.
For more info see: https://talks.cam.ac.uk/talk/index/231871
To join online, please contact yz506@cam.ac.uk
Refreshments will be provided after the talk for networking