Maya-7 CubeSat
Moving image analysis from the ground station into the satellite
An image classification hardware unit for the Maya-7 CubeSat, with a neural network running inside its microcontroller, so images are judged the moment they're captured instead of weeks later on the ground.
Who it's forTeams that need AI to make decisions on the device itself, where connectivity is slow, costly or absent.
- projected time to get full images down to Earth
- 2 → 1 week
- image judgement moved from the ground station into the satellite
- On-board
The challenge
A small satellite can only send so much data back to Earth. Every image had to be downlinked first and judged later at a ground station, which meant waiting about two weeks to receive full images, including the unusable ones.
What we built
An image classification hardware unit for the Maya-7 CubeSat, built as Master’s engineering research. A neural network runs directly inside the unit’s microcontroller and analyzes each image immediately upon capture. The judgement that used to happen on the ground now happens inside the satellite, so it can prioritize the images worth sending.
The result
- Full-image downlink time projected to drop from about two weeks to one.
- Better image quality and correctness expected, since poor images can be filtered before they use up bandwidth.
Both results are projections: the satellite has not yet launched, so they have not been tested in orbit.
Why it matters for your business
The same skills (fitting a model into tight hardware, making it reliable, and deciding locally instead of in the cloud) are what it takes to run AI offline, on-premise or on the factory floor, where sending data out is slow, costly or not allowed.
