Machine-Learned Satellite Segmentation for Unknown Hardware Configurations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation techniques for satellite components rely on prior knowledge of the satellite's hardware configuration, limiting their applicability to unknown or newly launched satellites.
Innovation Solution
A satellite classification system using machine learning and deep learning techniques generates image segmentation maps for satellites with unknown configurations, outputting pixel labels for different components and pose parameters without prior knowledge, employing a neural network architecture trained on diverse satellite imagery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image segmentation techniques are used for satellite components, then segmentation accuracy is improved when prior knowledge is available, but applicability deteriorates for unknown or newly launched satellites
Solution Approach 1:
The patent transforms the classification approach from rule-based (requiring prior knowledge of satellite parameters) to data-driven machine learning, where the model learns parameters automatically from training data. This allows the system to adapt to unknown satellite configurations by changing the fundamental parameter representation from fixed rules to learned features.
Solution Approach 2:
The patent uses synthetic satellite images generated from 3D models as training data, creating copies of real satellite appearances without requiring actual prior knowledge of specific satellites. This allows the model to learn from diverse synthetic examples and generalize to unknown real satellites, resolving the contradiction between accuracy and adaptability.
2Adaptability or versatility
If machine learning models are trained on diverse satellite imagery, then adaptability to unknown satellites is improved, but training data requirements and computational complexity increase
Solution Approach 1:
The patent generates synthetic training images by rendering 3D satellite models, creating unlimited diverse training data without requiring actual satellite images. This copying approach from 3D models to 2D images provides diverse training examples while avoiding the complexity of collecting and annotating real satellite imagery.
Solution Approach 2:
The patent performs preliminary generation of synthetic training data and pre-training of the model before actual satellite classification tasks. This preliminary action of creating training data and pre-training the model reduces the complexity and data requirements for subsequent deployment on unknown satellites.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for satellite component classification includes, at a satellite classification system, receiving a test image depicting a satellite, the satellite having a hardware component configuration that is unknown to the satellite classification system. The test image is input to a satellite classification model trained, based at least in part on a plurality of training satellite images, to generate output image segmentation maps for input satellite images. The satellite classification model outputs an output image segmentation map for the test image, one or more position parameters for the satellite, and one or more attitude parameters for the satellite. The output segmentation map includes a plurality of map pixels corresponding to a plurality of image pixels in the test image, wherein pixel values of the plurality of map pixels classify corresponding image pixels of the test image as depicting different hardware components of the satellite.