The aim of this presentation is to introduce cloud computing platforms that have now made it possible for us to have access to a wide variety of Satellite Imagery data along with the computing power required to run sophisticated Machine Learning models on that. Also, it briefly explains one available dataset and what problems are currently being tackled with that, to motivate the use of geospatial datasets in solving real-life problems.
PS:- It is our* class presentation given for a course taken under Prof Shashi Shekhar at the University of Minnesota.
* - Rahul Bhojwani(rahulbhojwani2003@gmail.com) and Kate Kuehl(katekuehl@gmail.com)
From idea to production in a day – Leveraging Azure ML and Streamlit to build...
Improving access to satellite imagery with Cloud computing
1. University of Minnesota-Twin Cities
Rahul Bhojwani(bhojw005@umn.edu)
Kate Kuehl(kuehl088@umn.edu)
Improving Access to Satellite Imagery
with Cloud Computing
2. Challenges in Remote Sensing:
● Data size - Expensive to download and process satellite data
● Data availability
● Timeliness of available data
● Data management and handling
● Computation
● Machine Learning friendly datasets
3. Challenges in Remote Sensing:
● Data size - Expensive to download and process satellite data
1 petabyte = 1,000,000,000,000,000 bytes OR 1,000 terabytes
4. What is “Earth on AWS”?
https://aws.amazon.com/earth/
● Run on Amazon Web Services’ servers
● Accessible through command line when logged into AWS account
● Used by MatLab, Mapbox, Esri, PlanetLabs, DigitalGlobe, NVIDIA, and more
5. What is “Google Earth Engine”?
https://earthengine.google.com/
● We used this in class
● Used by WRI, FAO, U of M, and other non-profits/universities
● Free for research, education, and nonprofit use
Python and JavaScript API’s available here: https://developers.google.com/earth-
engine/#api
8. Datasets available at AWS Earth
● SpaceNet Machine Learning Imagery
● National Agriculture Imagery Program
● GDELT - A Global Database of Society
● NASA Earth Exchange (NEX)
● DigitalGlobe Open Data Program
● UK Met Office Weather Forecasts
● Functional Map of the World
● Landsat
● NEXRAD
● Terrain Tiles
● GOES
● Sentinel-2
● OpenStreetMap
● MODIS
9. SpaceNet Machine Learning Imagery
● Commercial satellite imagery and labeled training data
● Intended for advancing innovation in Computer Vision
● To extract geometric features from remote sensing data such as:
○ Roads
○ Building footprints
○ Points of interest
14. Quiz
What challenges of remote sensing that’s addressed by cloud geospatial data?
A. Petabyte size of datasets
B. Timeliness of available data
C. Data management and handling
D. Computation
E. Machine Learning friendly datasets
15. Additional Sources
Information and Images
● Earth on AWS https://aws.amazon.com/earth/
● NDIVIA https://devblogs.nvidia.com/parallelforall/exploring-spacenet-dataset-using-digits/
● Google Earth Engine https://earthengine.google.com/
● NASA https://www.nasa.gov
● DigitalGlobe http://ow.ly/rIhV307X3u6
Images
● Wikipedia Commons https://commons.wikimedia.org/
Editor's Notes
“Big Geospatial Data – an OGC White Paper”
Open Geospatial Consortium
September 25, 2017
Ask the class if anyone has heard of itIt has existed in some form or another for 2 years