Determination of Satellite Solar panel and bus size from radar images with deep learning
conference paper
Radar systems play a very important role in Space Situational Awareness (SSA) since
they are able to provide a wealth of information on remote objects. In particular, imaging
radars can characterize space objects using Inverse Synthetic Aperture Radar (ISAR)
techniques. Under the right circumstances, ISAR images can provide details of a satellite
up to cm level or even lower. However, understanding the capabilities of an unknown satellite
from an ISAR image is not trivial. A first-order estimation of the satellite’s capabilities
can be determined simply from its size, particularly from its solar panel size. With this information,
parameters such as the satellite power budget can be inferred. In this paper, we
explore the use of Deep Learning (DL) techniques to determine the size and shape of the
solar panels and satellite bus from ISAR images. To provide a first analysis of the accuracy,
we compare DL output to the results given by a basic threshold-based technique. We
find that DL outperforms the thresholding approach, particularly when multiple classes are
distinguished from the background.
they are able to provide a wealth of information on remote objects. In particular, imaging
radars can characterize space objects using Inverse Synthetic Aperture Radar (ISAR)
techniques. Under the right circumstances, ISAR images can provide details of a satellite
up to cm level or even lower. However, understanding the capabilities of an unknown satellite
from an ISAR image is not trivial. A first-order estimation of the satellite’s capabilities
can be determined simply from its size, particularly from its solar panel size. With this information,
parameters such as the satellite power budget can be inferred. In this paper, we
explore the use of Deep Learning (DL) techniques to determine the size and shape of the
solar panels and satellite bus from ISAR images. To provide a first analysis of the accuracy,
we compare DL output to the results given by a basic threshold-based technique. We
find that DL outperforms the thresholding approach, particularly when multiple classes are
distinguished from the background.
TNO Identifier
981139
Source title
3rd IAA Conference on Space Situational Awareness (ICSSA) Madrid, Spain
Files
To receive the publication files, please send an e-mail request to TNO Repository.