Nantes Machine Learning Meetup

 
 
 

Multispectral Satellite data is plentiful, powerful and (mostly) free to obtain [1].
But where can we get it? How can we use it? What does it actually tell us?
How can ML and AI help us making sense of it?
In this talk I will outline the answer to these questions, sketching a
broad picture of the field of GeoDataScience and its numerous applications.
I will focus on the physical properties of the data and how they are used to monitor the environment [2, 3], natural disasters [4] and land use changes with time [5] as well as the data pre-processing and post processing and analysis techinques.
For the former, I will address some of the main problem that still plague data imagery analysis, such as those related to cloud coverage, and the proper normalization schemes for the images.
Regarding the latter, I will show how simple algorithms (such as the NDVI computation) as well as deep-learning based techniques (such as U-Net based semantic segmentation) are used to analyze satellite imagery.

Edoardo Carlesi was born in Rome, where he obtained his BSc/MS in Theoretical Physics. After a postdoc in Madrid and two postdocs in Jerusalem and Berlin, he moved back to Rome where he's working as Machine Learning Specialist for Top Network SpA and bass player for Nanowar Of Steel.

Ref.
[1] https://sentinels.copernicus.eu
[2] https://iforest.sisef.org/abstract/?id=ifor4043-015
[3] https://www.mdpi.com/2072-4292/13/21/4470
[4] https://plantsociology.arphahub.com/article/50519/element/2/16/
[5] https://www.nature.com/articles/s41598-020-74215-5