News
FAST-EO Press Release

FAST-EO Press Release

Copy URL to clipboard
Share on LinkedIn
Share on X
FAST-EO Press Release

FAST-EO project kicks off an ambitious AI geospatial foundation model initiative

 

The FAST-EO project, supported by the European Space Agency (ESA) Φ-lab, is dedicated to advancing Artificial Intelligence foundation models for Earth observation

 

Earth observation (EO) is about to get a major boost – thanks to FAST-EO, a new initiative funded by European Space Agency (ESA) Φ-lab and involving researchers from the German Aerospace Center (DLR), Forschungszentrum Jülich, KP Labs, and IBM Research. The project’s aim is to advance the integration and efficiency of multi-modal AI foundation models within the EO sector. Having kicked off on 5 February 2024, it is scheduled to run for 18 months, until mid-2025.

 

Typically done with the help of satellites, data from EO is crucial for sustainability and tackling the impacts of climate change. It is used for early-warning systems, monitoring of deforestation and afforestation, detection of illegal mining, forecasting floods, and much more.

 

As the amount of data we receive about Earth rises, multi-modal AI is pivotal to generate actionable insights at global scale to various stakeholders like emergency services, urban planners and decision makers in general.

 

To improve how the data is received and processed, these five organizations have decided to work together. Each partner brings unique strengths to the project, ranging from satellite data expertise to advanced AI and machine learning research, as well as high-performance computing, with the goal of enhancing the use of multi-modal AI foundation models within the EO community. By pooling their resources and expertise, the organizations hope to push the boundaries of what is currently possible in the field of AI for EO, making it more efficient, accessible, and powerful.

 

During the online kick-off meeting attended by all five partners, the coordinator of the project from the German Aerospace Centre (DLR), Dr. Ridvan Salih Kuzu, shared his views on the future transformative impact of this initiative on EO technologies. "By harnessing the power of AI foundation models, we're not just improving data processing and interpretation, but we're also paving the way for innovative applications in environmental monitoring, disaster response, and resource management," he said. "This project is a testament to the collaborative spirit of the European space sector and its commitment to innovation for the greater good of society and our planet."

 

In-depth discussions during the meeting centered on key aspects of the project, including advanced deep learning models, advanced machine learning techniques, relevant EO datasets critical to the initiative's success, and the efficient execution of various use cases.

The project aims to address several critical challenges:

 

  • Adapting AI models for EO data: One of the main objectives is to customise AI models to better handle the specific types of data encountered in EO applications. This involves tailoring models to process and interpret multi-satellite imagery and other relevant data forms effectively.
  • Enhancing multimodal data processing: FAST-EO seeks to improve the ability of AI models to process and analyse multimodal data, which includes a variety of data types such as images, text, and sensor data, among others. This capability is crucial for extracting meaningful insights from the diverse data sets used in EO.
  • Reducing costs: Another key focus is on reducing the computational and financial burdens associated with training and applying AI models. By making these processes more efficient, FAST-EO aims to make advanced EO technologies more accessible to a wider range of stakeholders.

 

Dr. Andrés Camero, co-head of the EO Data Science Department at DLR, expressed his enthusiasm about the potential of the project, and stated: "The kick-off of the FAST-EO initiative marks a significant milestone towards the establishment of European proprietary foundation models dedicated to open science and transparency. Moreover, this initiative holds outstanding importance in efficiently addressing societal challenges through EO, as it brings versatility and ease of access to AI adoption efforts in this field. At the DLR, our role will not only be to lead this initiative but also to bring the expertise of the remote sensing domain and facilitate the use of project outcomes in different fields. Ranging from forest monitoring to agricultural vulnerability assessment, we are aiming towards a more sustainable world through AI and EO."

 

Meanwhile, IBM Research, with collaborators, has developed Prithvi, one of the pioneering Earth observation foundation models of 2023. As part of IBM's commitment to open science, Prithvi is now available on HuggingFace. In the FAST-EO initiative, IBM experts will apply their expertise in pre-training, fine-tuning, and expanding EO foundation models. These models are designed to handle data from multiple ESA satellites and other modalities like text to unfold their full potential.

 

Dr. Juan Bernabe-Moreno, director of IBM Research UK & Ireland and strategy lead of Climate & Sustainability, said he was excited to extend IBM’s engagement to the European Space Agency. "Addressing global challenges like climate change requires truly global research efforts, which is at the heart of our open-science strategy," he stated. "Moreover, we must act quickly. Foundation models are crucial because they can be swiftly adapted to new climate challenges with minimal need for labour-intensive data labelling and are capable of generalizing effectively across different geographical regions, thereby democratising their advantages on a global scale."

 

Dr. Jakub Nalepa, head of AI at KP LABS, expressed his enthusiasm for their participation in the FAST-EO project. "We believe that the technology developed in this project will become a major step in unleashing the real potential of AI in EO, where we generate massive amounts of unlabelled data daily . FAST-EO can not only transform the way we process and ultimately benefit from multi-modal satellite data using large-capacity deep learning models, but also will it address societal and environmentally important EO tasks, making FAST-EO even more exciting."

Dr. Stefan Kesselheim, head of the Applied Machine Learning Lab at the Jülich Supercomputing Centre (Forschungszentrum Jülich), expressed the belief that "EO can make the world a better place, but achieving this requires both substantial brainpower and computational resources. The AI models of FAST-EO can be a great starting point for future EO activities that can lead to sustainable solutions for our planet. The Jülich Supercomputing Centre aims to provide the best infrastructure for such projects, and we are looking forward to seeing JUPITER in action, our upcoming first European Exascale supercomputer. We are delighted to be on board and are eager to contribute to advancing EO technologies that harness the potential of high-performance computing to address global challenges."

 

FAST-EO is an 18-month project (2024-2025) funded by ESA Φ-lab. "A cornerstone piece of the current strategy is the development and deployment of large-scale self-supervised EO foundation models. Numerous independent activities are converging in this domain, with FAST-EO being a particularly significant element of this overarching initiative," comments Nicolas Longépé, Earth observation data scientist at ESA Φ-lab. "Each partner brings unique strengths to the project, ranging from satellite data expertise to advanced AI and machine learning research, as well as high-performance computing, with the goal of enhancing the use of multi-modal AI foundation models within the EO community. By pooling their resources and expertise, we hope to push the boundaries of what is currently possible in the field of AI for EO, making it more efficient, accessible, and powerful."