2023-07-31 12:45:16
Data collected by remote sensing satellites is essential for many important activities, including aerial mapping, weather forecasting and deforestation monitoring. Currently, most satellites can only collect data passively, as they are not equipped to make decisions or detect relevant changes in what they observe. To detect changes and make decisions, the data must be transmitted to Earth to be processed there, which usually takes several hours or even days. This limits the ability to identify and react to rapidly occurring events, such as a natural disaster.
To overcome these constraints, a group of researchers, led by Vít Růžička from the University of Oxford in the UK, took up the challenge of training the first off-Earth machine learning program. Machine learning is a form of artificial intelligence.
In 2022, the team presented their idea to those responsible for the Dashing through the Stars mission, who had published a call for proposals for projects that could be carried out aboard the ION SCV004 satellite, launched in January 2022. The project was selected and, during the fall of 2022, the team sent the program code to the orbiting satellite.
The researchers trained a simple machine learning model for ION SCV004 to be able to directly detect changes in cloud cover from satellite images of Earth.
Illustration of the data used to train the tiny cloud classification model (left) and the predictions about the content of new images (right). (Image: Sentinel-2 data: ESA; processing: Vít Růžička)
The model was based on a strategy that allows it to learn the most important features to look for when there are only a few samples for training. A decisive help is that the data can be compressed into smaller representations, which makes the model faster and more efficient.
While the first part of the model, to compress the newly viewed images, was trained on the ground, the second part (which determined whether the image contained clouds or not) was trained directly on the satellite. (Source: NCYT from Amazings)
#equip #satellite #artificial #intelligence