BIG DIVE 8th edition

June 17th to July 12th 2018 | Turin, Italy

 

EIGHTH EDITION DIVERS LIST


/ STEFANO BELLINGERI
/ DAVIDE BELOCCHIO
/ FEDERICO BOMBARDIERI
/ RUBEN DE MARCH
/ THÉO DUPUIS
/ PIETRO FLORIO
/ MATTIA FOLCARELLI
/ JOSE FRANCISCO GUERRERO TELLO
/ THORSTEN KREISSIG
/ ELENA LABENETS
/ LUCIA MARINOZZI
/ YAMATO MINAMI
/ NICOLAS MONVERT
/ NICOLA OCCELLI
/ DAVIDE PRATI
/ VALENTINA TAMBURELLO

TEACHERS & SPEAKERS


Fabio Franchino – todo.to.it
Stefania Delprete – top-ix.org
Maurizio Napolitano – fbk.eu
Alessandro Molina – axant.it
Niccolò Bidotti – agilelab.it
Francesco Tarasconi – CELI
Alex Comunian – ThoughtWorks
Leif Toudal Pedersen  – TUD
Cristiano De Nobili – Harman-Samsung
Andrè Panisson, Alan Perotti – isi.it
Matteo Chinazzi – MoBS Lab

Guest speakers
Pierre-Philippe Mathieu – ESA
Stefano T. Chiadò – Vastalla
Marco Aldinucci – University of Turin
Antonio Vetrò – Politecnico di Torino

Demo
Cristiano Nattero, Paolo Campanella – WASDI
Grega Milcinski – Sentinel Hub
Juan B. Pedro – Starlab

DATA SPONSORS


ALTEC
altecspace.it/en

ESA
esa.int/ESA

INRIM
inrim.eu

 

TECH PARTNERS AND STAKEHOLDERS


ARPEA
arpea.piemonte.it

Sentinel Hub
sentinel-hub.com

Technical University of Denmark
dtu.dk/English

University of Turin
en.unito.it

FINAL PROJECTS

“Atomic clocks and Space Weather”
Data by: INRiM

Team Group: T. Dupuis, N. Monvert, D. Belocchio, S. Bellingeri,
Team Composition: Developers: x2 | Data Scientist: x1 | Domain Expert: x1

Could space weather events affect the frequencies of the atomic clocks of GNSS satellites?
To answer this question, INRiM provided a dataset containing the frequencies of the atomic clocks in 17 Galileo satellites from June 2016 to March 2018. The group collected and processed additional data on the earth magnetic field and weather events measured in GOES satellites. These datasets were used to detect and validate geomagnetic storms and identify potential correlations with clocks anomalies.

Additionally, they designed an RNN and trained it on a subset of the dataset provided by INRiM to detect anomalies in the clock frequencies.

Tools: Python (BeautifulSoup, Bokeh, Keras, Pandas, Seaborn, Sklearn), Jupyter Notebooks
Data Science Methods: RNN (LSTM)

“Polar: ice detection”
Data by: ESA, TUD

Team Group: L. Marinozzi, F. Bombardieri, M. Folcarelli, J. F. Guerrero Tello
Team Composition: Data Scientists: x2 | Developers: x2

The question behind the project work: Use a set of geospatial raster-images and other data collected by Sentinel-1 satellite to create a ML system for the detection of ice presence and concentration in order to automatically produce an ice map. The group explored the concentration of ice using The ASIP sea ice data set of 26 files in NetCDF format.

From that data they extracted matrices and transformed them in order to feed the machine learning model. The trained a U-net model and evaluate the quality of the learning by using the Jaccard coefficient.

Tools: Python, Keras, Tensorflow, QGIS, Jupyter Notebook
Data Science Methods: Image segmentation with U-Net model

“Earth Observation and Agriculture”
Data by: ALTEC Space, ARPEA

Team Group: R. De March, N. Occelli, E. Labenets, Y. Minami
Team Composition: Domain Expert: x1 | Data Scientist: x1 | Developers: x2

Is it possible to automate the identification of the various farming types, using EO images and a CNN Deep Learning model? This is the question they had to answer to ALTEC.

The work path started by processing the EO images coming from Sentinel 1 and 2 satellites (Pre-processing chain already applied by ALTEC). They were then feeded to a FCN U-Net to perform image segmentation in order to classify crops.

Tools: Python and libraries, Jupyter Notebook, Keras framework, Rasterio, QGIS, SNAP, MS Excel
Data Science Methods: Fully Convolutional Neural Network, Image Segmentation

“Earth Observation and Urbanization”
Data by: ESA, SENTINEL HUB

EARTH OBSERVATION AND URBANIZATION

Team Group: V. Tamburello, P. Florio, D. Prati, T. Kreissig
Team Composition: Domain Experts: x2 | Developer: x1 | Researcher: x1 |

The question behind the project work: starting with Sentinel 2 satellite images of a city, is it possible not only to distinguish between urbanised and non-urbanised areas, but also to predict the future urbanisation?

The team used satellite images of Las Vegas and Manaus from 2015. From these random patches have been extracted in order to train a U-net model, able to classify urban and non-urban areas. Predictions for 2019 have been made with the same algorithm, allowing the team to estimate a growth rate of 14% for Manaus and % for Las Vegas. Comparisons with 2019 satellite images of the two cities showed a good match.

Tools: Jupyter Notebook, Python, Keras, ArcGIS
Data Science Methods: Fully Convolutional U-Net