Semantic Segmentation of Low Earth Orbit Satellites Using Convolutional Neural Networks
Abstract
Semantic segmentation is taking an image that has different components, like sky, tree, cat, cow, etc. and identifying which parts of the image belong to each component. For this study, we wanted to see if we could transfer this task over to images of satellites to identify which parts of the images were the different components of the satellite such as bus, solar panel, antenna, payload, etc. This can be very helpful for a number of applications including providing health and status of a satellite, or understanding its behavior. With the pristine images, this might not be too hard. By eye, we can roughly produce a segmentation, but with turbulent images, this becomes much, much harder. And for either turbulent or pristine images, semantics segmentation becomes very tedious with large volumes of images. However, semantic segmentation of images in other fields such as biology and even day-to-day applications has seen revolutionary improvements through convolutional neural networks, a machine learning technique. So we wanted to apply CNN to semantic segmentation of satellites as well.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 08, 2022
- Accession Number
- AD1204607
Entities
People
- Andrew Van Berg
- Jacob Lucas
- Julia Yang
- Justin Fletcher
- Michael Abercrombie
- Trent Kyono
Organizations
- Air Force Research Laboratory
- Boeing
- United States Space Force