GeoUAV Keynote 1
Aligning data representations of remote sensing multitemporal images
Abstract: The possibility of re-using labeled data when new image acquisitions become available is very appealing for modern remote sensing analysts. Data acquisition has become faster and several sensors with similar characteristics can be used to solve monitoring tasks. Such a re-use of terrain data would allow to transfer models, which are optimal to describe a region imaged at a certain time instant, to new image acquisitions observing neighboring regions and/or different time instants. Currently, re-using field data (and corresponding trained classifiers) is hampered by the failure of such models to deal with changes in either the acquisition conditions or differences in the dimensionality of the data spaces. These changes impact the shape and structure of the space where the data live, their manifold and such differences have catastrophic consequences in model transferrability.
During this talk, I will present recent methodologies to align data manifolds. These methodologies allow to define projections to a common data space where a single model can be used to process a series of remote sensing images taken under different conditions and/or by different sensors. Solutions to reduce the level of supervision in the definition of the projections will also be discussed.
Bio: Devis Tuia was born in Mendrisio, Switzerland, in 1980. He received a diploma in Geography at the University of Lausanne (UNIL) in 2004, the Master of Advanced Studies in Environmental Engineering at the Federal Institute of Technology of Lausanne (EPFL) in 2005 and a Ph.D. in Environmental Sciences at UNIL in 2009. He was then a visiting postdoc researcher at the University of Valencia, Spain and the University of Colorado, Boulder, CO. He then worked as Senior Research Associate at EPFL under a Swiss National Foundation program. Since 2014, he is Assistant Professor at the Remote Sensing Laboratories, Department of Geography of the University of Zurich. His research interests include the development of algorithms for information extraction and data fusion of remote sensing images using machine learning and vision algorithms. Dr. Tuia serves as a Co-chair of the Image Analysis and Data Fusion Technical Committee of the IEEE GRSS. He is an Associate Editor of the IEEE Journal of Selected Topics in Applied Earth Observation and Remote Sensing (IEEE JSTARS). Visit http://devis.tuia.googlepages.com/ for more information.