Global and Dense Embeddings of Earth: Major TOM Floating in the Latent Space
Mikolaj Czerkawski, Marcin Kluczek, Jędrzej S. Bojanowski
2024-12-10
Summary
This paper talks about a new method for creating efficient representations of Earth observation data using a project called Major TOM, which helps make satellite imagery easier to analyze and understand.
What's the problem?
With the increasing amount of data collected from satellites, there is a need for better ways to represent this information. Current methods often rely on limited data and do not effectively capture the details needed for accurate analysis, making it difficult to extract meaningful insights from the vast amounts of geospatial data available.
What's the solution?
The authors propose an extension to the Major TOM project that focuses on creating global and dense embeddings of Earth observation data. They developed a new dataset that includes high-quality representations of geospatial visual data, which can be used by AI models for analysis. This dataset is open and free, allowing researchers and developers to access and utilize it without restrictions. The embeddings help simplify complex satellite images into more manageable forms while preserving important information about the Earth's surface.
Why it matters?
This research is important because it enhances how scientists and analysts can work with satellite data, making it easier to study environmental changes, urban development, and other geographic phenomena. By providing a comprehensive and accessible dataset, this work supports better decision-making and research in fields like climate science, urban planning, and disaster management.
Abstract
With the ever-increasing volumes of the Earth observation data present in the archives of large programmes such as Copernicus, there is a growing need for efficient vector representations of the underlying raw data. The approach of extracting feature representations from pretrained deep neural networks is a powerful approach that can provide semantic abstractions of the input data. However, the way this is done for imagery archives containing geospatial data has not yet been defined. In this work, an extension is proposed to an existing community project, Major TOM, focused on the provision and standardization of open and free AI-ready datasets for Earth observation. Furthermore, four global and dense embedding datasets are released openly and for free along with the publication of this manuscript, resulting in the most comprehensive global open dataset of geospatial visual embeddings in terms of covered Earth's surface.