Speaker: Mihai Datcu (IEEE Fellow)

Affiliation: POLITEHNICA Bucharest

Report Title: Physics Aware AI for Synthetic Aperture Radar Earth Observation

Abstract: 

The presentation will address methods for learning from the image formation, embedding domain knowledge explaining causalities and interpreting the objects, their signatures, and physical parameters. The new paradigm is centered on a systematic representation of the SAR imaging.  Thus, the SAR data information extraction is based on the model of the physical layers in the SAR systems: i) Orbits & image formation; ii) Sensor model, iii) EM radiation-scene interaction model. The methods are applied to implement Virtual Sensing functionalities, i.e. the prediction of un-observed SAR wavelengths or polarization.  The dedicated methods are beyond the complex value deep learning models. The new architectures are learning and preserving the basic SAR properties, as the coherence and the azimuth subaperture characteristics. The addressed methods cover complex valued DNN architectures for SLC SAR data classification embedding SAR signal processing. The methods are demonstrated for scene signatures prediction, addressing the interaction between the EM radiation and the Earth surface. Further a new category of physics aware generative models is introduced. Derived from autoencoders, GAN or diffusion the new models generate SAR images beyond the realistic aspects, preserving their class identity or reproducing the underlying physical models. Examples will be provided for SAR applications for monitoring the effects of climate changes.

Biography: 

Mihai Datcu (Fellow, IEEE) received the M.S. and  Ph.D. degrees in electronics and telecommunications from the University Politehnica of Bucharest (UPB), in 1978 and 1986, respectively, and the Habilitation a Diriger des Recherches degree in computer science from University Louis Pasteur, Strasbourg, in 1999. Since 1981, he has been with the Faculty of Electronics, Telecommunications and Information Technology, POLITEHNICA Bucharest. From 1992 to 2002, he had an Invited Professor Assignment with the Institute of Communication Technology, Swiss Federal Institute of Technology (ETH Zürich). From 1993 to 2023, he was with the German Aerospace Center (DLR), Oberpfaffafenhofen, Germany, as a Senior Scientist with the Remote Sensing Technology Institute (IMF) and a Team Leader of the Big Data and AI for Earth Observation. From 2005 to 2013, he was a Professor holder of the DLR-CNES Chair, ParisTech. From 2018 to 2020, he was the holder of the Blaise Pascal International Chair of Excellence, at Conservatoire National des Arts et Métiers (CNAM), Paris. Presently he is Visiting Professor with the ESA’s Φ-Lab. His research interests include information theory, signal processing, artificial intelligence, computational imaging, and quantum machine learning with applications in EO. He was a recipient of the Chaire d’Excellence Internationale Blaise Pascal 2017 for international recognition in the field of data science in EO, and the 2018 Ad Astra Award for Excellence in Science. In 2022, he received the IEEE GRSS David Landgrebe Award in recognition of outstanding contributions to Earth observation analysis using innovative concepts for big data analysis, image mining, machine learning, smart sensors, and quantum resources. From 2020 to 2024 he was IEEE GRSS DL.