Detecting critical minerals from space: A machine learning approach to satellite imagery analysis.
Connection to one of the themes:
We are relevant to the theme of supporting sustainable infrastructure development because critical minerals are needed in almost every sustainable and green energy solution.
The problem we are solving:
Critical minerals tend to be expensive and time-consuming to detect, our solution cuts down substantially both on cost and time increasing the critical minerals availability in the market and thus supporting sustainable infrastructure development.
The data used for our training features:
We used Sentinel 2 L2A data from Copernicus using Open Earth Observation (openEO) API.
The difference between L1C and L2A images is that L2A has been atmospherically corrected.
From there, we took the three bands what is usually known as “the geology bands”:
12 - Short Wave Infrared, Central Wavelength 2190nm, resolution 20m
11 - Short Wave Infrared, Central Wavelength 1610nm, resolution 20m
02 - Visible: Blue, Central Wavelength 490nm, resolution 10m
…and a fourth band, the Scene Classification Layer (SCL) which helps to indicate whether a pixel is covered by a (type of) cloud or not. We used this SCL to build a cloud mask and applied it to the data.
After downloading the three bands, we saved it as a png, scaling and downsampling the measurement values to 0-255 range.
Resolution of this png image: 11758w, 5976h. These three bands internalised inside a png image were used as features inside a deep neural network.
Kristjan Roosild: Data science, software engineering, GIS (ex Wise, Pactum)
Arnel Pallo: Data science, software engineering (Pactum)
Mehdi Boubnan: Data Science (YC W21 Mindee)
Bouya Maalainine: Product, Sales, Management, Marketing, Finance (ex Wise)
Anderson Mahison Senior Geologist
El Khalifa Beyah: Ex Head of SNIM (second largest exporter of iron in Africa)