Cédric Pradalier has been appointed at GeorgiaTech Lorraine in September 2012 with the objective to extend the activity of the CNRS UMI 2958 GT-CNRS towards computational perception and robotics. He is the coordinator of the H2020 project BugWright2 on robotics for inspection and maintenance of large metallic structures and has been involved in H2020 project Flourish and FP7 project Noptilus, as well as in projects on environmental monitoring. In particular, the results presented in this talk have been acquired in the context of a France-Israel collaboration project within the Maimonide 2019 framework.
He brought over his experience on autonomous systems, and inspection robotics gathered at the Autonomous Systems Lab at ETH Zürich, Switzerland and the Autonomous Systems Lab at CSIRO, Australia. From November 2007 to August 2012, Dr. C. Pradalier has been deputy director of the Autonomous Systems Lab at ETH Zürich, in collaboration with Prof. Roland Siegwart. From 2004 to 2007, Dr. C. Pradalier was a research scientist at CSIRO, Australia. He was then involved in the development of software for autonomous large industrial robots and an autonomous underwater vehicle for the monitoring of the Great Barrier Reef, Australia. He received his Ph.D. in 2004 from the National Polytechnic Institute of Grenoble (INPG) on the topic of autonomous navigation of a small urban mobility system.
In the context of smart irrigation systems, modeling the evapotranspiration of crops is of paramount importance. However, when it comes to filling gaps of missing data in evapotranspiration measurements, this field still heavily relies on techniques based on a marginal distribution sampling. Similarly, to estimate the evapotranspiration of the crops in real-time the prevalent technique is based on the Penman-Monteith equation coupled to a correction coefficient. This led us to investigate whether Deep Learning could be better suited to model evapotranspiration than classical approaches?
After all, given a large enough dataset, data-driven methods could extract pattern that are specific to a locality, a crop, a season. In this talk, we explore two main tracks: first, we discuss multi-head deep attention networks and show how they can be applied to fill gaps in evapotranspiration measurements. This methodology couples large-scale information extraction with modeling capabilities that cannot be achieved by interpolation-like techniques. We demonstrate their performance on real-data and compare their performances to REddyProc. In a second step, we demonstrate how neural-networks can be used to perform real-time estimation of crop evapotranspiration at either a daily rate or half-hourly rate.
Authors: Cédric Pradalier, Antoine Richard (GeorgiaTech Lorraine UMI2958 GT-CNRS, France), Lior Fine (Institute of Soil, Water and Environmental Sciences, Agricultural Research Organization,Volcani Center, Israel/The Robert H Smith Faculty of Agriculture, Food and Environment, The Hebrew University of Jerusalem), Josef Tanny(Institute of Soil, Water and Environmental Sciences, Agricultural Research Organization,Volcani Center, Israel/HIT – Holon Institute of Technology, Holon, Israel), Offer Rozenstein (Institute of Soil, Water and Environmental Sciences, Agricultural Research Organization,Volcani Center, Israel), Matthieu Geist (Google Research Brain Team)
Ask the Speaker
You can post your questions for the speaker of this talk here until Nov 8. Your questions are visible to all other registered participants of the conference and will be visible together with your username that consists of the beginning of your email address. Participants can “like” your question. Questions with more like are more likely to be discussed during the live Q&A session on November 10, 2020.