14. Dezember 2023
Self-Supervised Learning for SAR; Benchmarking Datasets and Performance on Diverse Downstream Tasks
ESA Climate Office Research Fellow, Dr Anna Jungbluth, presents on 16 Dec at AGU2023
ESA climate office Research Fellow, Anna Jungbluth will discuss the potential – and benchmarking results – related to the use of machine learning techniques to leverage Earth observations, specifically SAR data at the AGU conference on 16 Dec (16:10 PM PST 01:10 CEST) - West (Level 2, West, MC).
Dr Jungbluth’s presentation explores how pre-trained machine learning models perform on unseen satellite data to assess the generalisability of the approach in the context of exploiting SAR data.
Machine learning models that learn features in the data without being provided with explicit labels offer great potential to fully leverage the wealth and complexity of the available data.
Pre-trained models have significant potential to support a variety of downstream tasks, such as prediction of local fire and flood events and changes in annual land characteristics including vegetation percentage, land cover, above-ground biomass.
In her presentation, Dr Jungbluth provides examples of global benchmarking datasets of SAR input data and associated downstream tasks ready for use in machine learning pipelines. The data to be shown contains contains ~500,000 co-registered and aligned tiles, covering South America, the US, Europe, and Asia.
Dr Jungbluth is a Research Fellow in the Climate Office of the European Space Agency (ESA). Her research is focused on applying machine learning techniques to ESA's satellite-derived climate data.
The work to be presented has been developed in collaboration with ESA’s Climate Office and ESA Φ-lab, the latter focussed innovations to accelerate Earth Observation (EO), partners of theFrontier Development Lab - a public-private partnership AI research initiative - and the University of Oxford.