Earth Observation data analysis plays a significant role in understanding our planet and its dynamics. Indeed, satellite images are acquired worldwide, with temporal revisit every few days and a spatial resolution of a few meters. Satellite programs provide the means for global monitoring from afar, enabling activities such as studying climate change effects, building resilient ecosystems, and reacting to disasters in a timely manner. However, accessing the information contained in EO data remains challenging: the quantities of data are significant, and the information extraction process remains behind complex data processing streams, making it accessible only to experts.
This project aims to develop approaches to describe satellite images and their changes using natural language, yielding explanations that are accessible to everyone. These descriptions can then be used to explain processes at work in the images or to answer user-related questions. Together with the Natural Language Processing (NLP) lab, we work on developing machine learning approaches to generate explanations of EO imagery by exploiting the knowledge available in large textual corpora such as Wikipedia or specialized webpages.