Star Tracker Using Deep Learning for Low-Cost Attitude Determination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current star trackers are limited by high costs due to the need for expensive, high-precision detectors and optics to achieve moderate accuracy, restricting their use to systems with large budgets, and they are susceptible to performance degradation in space environments.
Innovation Solution
The use of vector-based deep learning, specifically with Hinton's capsule networks or coordinate convolution layers, to classify and localize features in full-frame images of stars, even when the image is defocused, allowing for attitude determination with commercial off-the-shelf, low-cost CMOS detectors and optics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If expensive, high-precision detectors and optics are used to achieve moderate accuracy, then measurement precision is improved, but device cost increases
Solution Approach 1:
The patent replaces expensive, high-precision detectors and optics with commercial off-the-shelf (COTS) components that are lower cost and have reduced precision. The deep learning algorithm compensates for the lower quality of these inexpensive components, achieving comparable attitude determination accuracy without requiring high-precision hardware
Solution Approach 2:
The patent changes the operational parameters of the imaging system by using defocused images instead of focused images. This parameter change allows the use of lower-cost optics while the deep learning algorithm processes the defocused images to extract star positions and determine attitude, transforming a traditionally harmful effect (defocus blur) into a useful feature
2Ease of manufacture
If commercial off-the-shelf, low-cost CMOS detectors and optics are used, then device cost is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent introduces a deep learning algorithm as an intermediary between the low-cost CMOS detector/optics and the attitude determination process. This intermediary processes the images from the inexpensive components, extracting star positions and attitudes with accuracy comparable to systems using expensive high-precision hardware
Solution Approach 2:
The patent replaces the mechanical/optical precision requirements with a computational approach. Instead of relying on precisely engineered optics and detectors, the system uses a deep learning algorithm that processes images computationally, substituting mechanical precision with algorithmic intelligence
3Ease of manufacture
If defocused images are used with low-cost optics, then device cost is reduced, but image quality deteriorates
Solution Approach 1:
The patent converts the harmful effect of defocus blur into a beneficial feature. By intentionally using defocused images from low-cost optics, the system creates a consistent image characteristic that the deep learning algorithm is trained to recognize and process, transforming what would normally be considered image degradation into a useful input pattern
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach significantly reduces manufacturing, test, and material costs while maintaining attitude accuracy comparable to standard moderate accuracy star trackers, using the additional features from blurred images to compensate for lower quality optics and electronics.
Implementation Method 1
an uncooled infrared detector array... the optical system and detector can be arranged such that the image information is blurred or defocused as it is received at the focal plane of the detector
Data Source
AI summary
Star tracker systems and methods are provided. The star tracker incorporates deep learning processes in combination with relatively low cost hardware components to provide moderate (e.g. ˜1 arc second attitude uncertainty) accuracy. The neural network implementing the deep learning processes can include a Hinton's capsule network or a coordinate convolution layer to maintain spatial relationships between features in images encompassing a plurality of features. The hardware components can be configured to collect a blurred or defocused image in which point sources of light appear as blurs, and in which the blurs create points of intersection. Alternatively or in addition, a blurred or defocused image can be created using processes implemented as part of application programming. The processing of collected images by a neural network to provide an attitude determination can include analyzing a plurality of blurs and blur intersections across an entire frame of image data.