GEODE · GEODE: GEographical Object Detection, Explained

The GEODE project brings explainability to deep learning for Earth observation. GEODE aims at improving the transparency of computer vision models used for analyzing satellite imagery in large-scale mapping projects. As AI models are increasingly used in applications such as environmental monitoring and land-use classification, their opaque decision-making processes limit trust and usability. Using novel XAI (eXplainable AI) techniques, GEODE will offer pixel-level explanations to understand why a given model fails or succeeds on certain objects. Performance maps will indicate where the model is expected to have standard performance. Finally, an active learning strategy will be developed, to define how to conduct labeling work for global mapping.Answering the why, where and how deep mapping models work, GEODE is a bold step towards unlocking the full potential of AI paired with large volumes of satellite imagery. The project will take place at the Image & Signal Processing Laboratory at Universitat de València in Professor Gustau Camps-Valls' group, a world-renowned leader in machine learning for Earth observation.