New B2B commerce study
Download now
Aerospace, Machine Learning
Detecting Clouds with Machine Learning
New satellite imagery solutions based on existing technology
Providing value on Earth and beyond
Where human intelligence meets volume boundaries, Machine Learning starts. This was the case for the issue we needed to solve in this project: making sure that clouds do not interfere with the results we can extract from satellite imagery.
This project was carried out in a partnership of ZAMG and us.
CI4Clouds was funded by the Austrian Federal Ministry of Transport, Innovation and Technology (BMVIT, meanwhile BMK) under the program “ICT of the Future” between 2015 and 2016.
The challenge
A cloudy problem
As trivial of an issue as it may seem, 55% of the Earth surface is always clouded. This is a problem when it comes to acquiring Earth Observation imagery: once massive amounts of image data come in, it becomes necessary to leave out the pixels corresponding to clouds and keep those in which the Earth surface is visible.
Misclassification has a strong impact on further algorithms on a processing chain, which assume clear sky conditions. A solution is built into some satellites that equip instruments dedicated to detecting clouds.
However, not all satellites have it, yet they produce data, too: solving this is a big challenge if one wants this data not to be flawed.
The solution
CI4Clouds
We set up against each other several state-of-the-art machine learning algorithms. These are Deep Learning (DL) / Convolutional Neuronal Nets (CNN) and Random Forests (RF) based approaches as well as Support Vector Machines (SVM).
For training them we used high-performance hardware and carried out the evaluation against existing cloud masks: Tree-based approaches turned out to be the winning algorithms.
Random Forests performed better than the known cloud-masking products, resulting in us being able to use cloudmask to get the same pre-processing results as with traditional cloud mask products.
We applied this same approach to our project BeatIt, where we use the cloudmask to detect and forecast the bark beetle infestations in forests, thus helping local stakeholders in protecting them.
The main lesson we learned here is one: sky is not the limit, and you can always go beyond even the thickest of clouds. If you want to fulfill your digital potential or are wonder if your problem can be solved with Machine Learning, reach out to us!
Austrian Federal Ministry of Transport
This project was carried out in a partnership of ZAMG and Cloudflight. CI4Clouds was funded by the Austrian Federal Ministry of Transport, Innovation and Technology (BMVIT, meanwhile BMK) under the program “ICT of the Future” between 2015 and 2016.
