My research is in geospatial computer vision, a field at the interface between GIscience, remote sensing and machine learning. I develop digital solutions to address problems of land planning and the environment. I led most of my efforts in urban recognition, land use modeling and analysis, but I also have experience in wildlife tracking, environmental risk reduction and forest management through scientific collaborations.
During the last five years, I have focused on the development of methods for urban analysis that can make sense of multi-sensorial data, e.g. remote sensing image data, social media, terrestrial campaigns and all kind of open access data. I design advanced machine learning algorithms, rooted mainly in deep and structured outputs learning. Such learning paradigms allow me to mine large datasets, while injecting prior knowledge about the problem of interest. I also devoted wide efforts in the question of adaptation of machine learning to evolving acquisition conditions, in order to increase the generality and robustness of urban recognition methods and make them more effective in multi-cities environments.
I work closely with application experts, in order to stay close to the field’s expertise and avoiding forcing the algorithm to re-learn structure that already known and that could be injected efficiently. I have worked intensively towards interactive systems exploiting human-machine interaction and applied mainly in problems of remote sensing image analysis involving human annotators.