Revolutionary AI Model Trained in Space Dramatically Expedites Satellite-Based Data Analysis
Satellites play a crucial role in various sectors, including aerial mapping, weather forecasting, and deforestation monitoring. However, the existing systems are primarily passive data collectors, lacking the ability to make decisions or detect changes in real-time. Traditionally, the collected data needs to be transmitted to Earth for processing, which can take hours or even days, causing delays in responding to emergent situations such as natural disasters.
To address this issue, a research team from the University of Oxford’s Department of Computer Science, led by DPhil student Vit Ruzicka, embarked on a groundbreaking project to train a machine learning program in space. Their proposal was selected in 2022 by the Dashing through the Stars mission, following an open call for projects to be executed aboard the ION SCV004 satellite, launched earlier that year. The team successfully uploaded the program’s code to the satellite already in orbit.
The innovative model, known as RaVAEN, was specifically trained to identify changes in cloud cover using aerial images taken directly onboard the satellite. This approach deviates from the conventional method of training models on Earth. Using a few-shot learning technique, the model learns key features from a limited number of samples, compressing the data into smaller representations for increased speed and efficiency.
Explaining how RaVAEN works, Ruzicka stated, The model compresses large image files into 128-number vectors. During the training phase, it learns to retain only the informative values in this vector, those that correlate with the change it is trying to detect—in this case, the presence of clouds. As a result, the training process is remarkably fast due to the small size of the classification model.
While the initial part of the model was trained on Earth, the subsequent part responsible for detecting cloud presence was trained directly on the satellite. Unlike the typical development process of machine learning models involving multiple rounds of training with a cluster of computers, this compact model completed the training phase using over 1300 images in just about one and a half seconds.
In tests with new data, the model was able to automatically detect the presence or absence of a cloud within a tenth of a second. It efficiently encoded and analyzed a scene approximately equivalent to an area of around 4.8×4.8 square kilometers, which is nearly 450 football fields.
Ruzicka emphasized the adaptability of the model for various tasks and data types. The team plans to develop advanced models capable of distinguishing between significant changes, such as flooding, fires, and deforestation, as well as natural alterations like seasonal leaf color changes. Additionally, they aim to create models for more complex data, including hyperspectral satellite images, which could facilitate the detection of methane leaks, a vital step in combating climate change.
The integration of machine learning in space-based satellite systems could also mitigate issues with on-board sensors that often require recalibration due to harsh environmental conditions. Ruzicka noted, Our proposed system could be utilized in constellations of non-homogeneous satellites, where reliable information from one satellite can be applied to train the rest of the constellation.
Professor Andrew Markham, Vit’s DPhil research supervisor, believes that machine learning has immense potential to enhance remote sensing by making space-based data acquisition more autonomous and reducing the response time between data acquisition and action.
This impressive project serves as a proof-of-principle and was a joint venture between the University of Oxford, the European Space Agency (ESA) F-lab via the Cognitive Cloud Computing in Space (3CS) campaign, the Trillium Technologies initiative Networked Intelligence in Space (NIO.space), and partners at D-Orbit and Unibap.
In conclusion, the training of a machine learning model in space has revolutionized satellite-based data analysis. RaVAEN’s ability to quickly analyze data directly on the satellite enables near-instantaneous decision-making, enhancing the effectiveness of satellite systems in various sectors. With further advancements and adaptations, this technology holds immense potential for addressing global challenges such as climate change, disaster response, and environmental monitoring.
