A Digital Assistant for Digital Twin Earth

We present an AI-powered digital assistant that includes four search engines for satellite images (search by image, search by caption, visual question answering and knowledge graph question answering) that are orchestrated by a task interpreter in order to answer complex requests of users looking for Earth observation data. The digital assistant will be demonstrated in three use cases: vessel detection, water bodies dynamics and training dataset construction. The digital assistant builds on recent work of the academic project partners on deep learning techniques for satellite images, search engines for satellite images, visual question answering, question answering over knowledge graphs and linked geospatial data, and question answering engines for satellite data archives. This work is funded by the European Space Agency project "DA4DTE: Demonstrator Precursor Digital Assistant Interface For Digital Twin Earth''. The project consortium is led by the Italian company eGEOS, with the National and Kapodistrian University of Athens and the TU Berlin as subcontractors.