Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HS-ANET: Star Spectral Type Enhanced Astrometric Calibration for Hyper Spectral Space Imaging

Kevin Phan, William Mitchell, David Chaparro, Enrique De Alba, J. Zachary Gazak

TL;DR

This work addresses the difficulty of solving spacecraft astrometry under limited-star conditions by extending Astrometry.net with stellar spectral-type information derived from hyperspectral imagery. By adding a spectral verification step to the Bayesian decision framework, HS-ANET leverages spectral consistency to tighten hypothesis scoring and reduce reliance on geometric constraints alone. Evaluated on 958 Gaia DR3-based simulated skyfields, HS-ANET achieves a normalized AUC of $0.669$ in sparse regimes versus $0.361$ for the baseline, and remains effective with as few as five detected stars. The approach offers a practical path to more robust onboard star identification and space traffic management, enabling reliable localization in narrow-field observations.

Abstract

Traditional lost-in-space algorithms, such as those implemented in astrometry.net, solve for spacecraft orientation by matching observed star fields to celestial catalogs using geometric asterisms alone. In this work, we propose a novel extension to astrometry.net that incorporates stellar spectral type, which is derived from hyperspectral imagery, into the matching process. By adding this spectral dimension to each star detection, we constrain the search space and improve match specificity, enabling successful astrometric solutions with significantly fewer stars. Our modified pipeline demonstrates improved fit rates and reduced failure cases in cluttered or ambiguous star fields, which is especially critical for autonomous space situational awareness and traffic management. Our results suggest that modest spectral resolution, when incorporated into existing geometric frameworks, can dramatically improve robustness and efficiency in onboard star identification systems.

HS-ANET: Star Spectral Type Enhanced Astrometric Calibration for Hyper Spectral Space Imaging

TL;DR

This work addresses the difficulty of solving spacecraft astrometry under limited-star conditions by extending Astrometry.net with stellar spectral-type information derived from hyperspectral imagery. By adding a spectral verification step to the Bayesian decision framework, HS-ANET leverages spectral consistency to tighten hypothesis scoring and reduce reliance on geometric constraints alone. Evaluated on 958 Gaia DR3-based simulated skyfields, HS-ANET achieves a normalized AUC of in sparse regimes versus for the baseline, and remains effective with as few as five detected stars. The approach offers a practical path to more robust onboard star identification and space traffic management, enabling reliable localization in narrow-field observations.

Abstract

Traditional lost-in-space algorithms, such as those implemented in astrometry.net, solve for spacecraft orientation by matching observed star fields to celestial catalogs using geometric asterisms alone. In this work, we propose a novel extension to astrometry.net that incorporates stellar spectral type, which is derived from hyperspectral imagery, into the matching process. By adding this spectral dimension to each star detection, we constrain the search space and improve match specificity, enabling successful astrometric solutions with significantly fewer stars. Our modified pipeline demonstrates improved fit rates and reduced failure cases in cluttered or ambiguous star fields, which is especially critical for autonomous space situational awareness and traffic management. Our results suggest that modest spectral resolution, when incorporated into existing geometric frameworks, can dramatically improve robustness and efficiency in onboard star identification systems.

Paper Structure

This paper contains 15 sections, 9 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Data Simulation and Sourcing
  3. Star Catalogs
  4. SatSim Data
  5. Methodology
  6. Source Extraction
  7. Quad Generation
  8. Hypotheses Generation
  9. Positional Verification (Bayesian Decision Process)
  10. Spectral Verification
  11. Results
  12. Fit Rate Analysis
  13. Normalized AUC Comparison
  14. Discussion
  15. Conclusion

Figures (3)

  • Figure 1: Proposed pipeline in which HS-ANET can be used to optimize low field of view telescope astrometric fitting.
  • Figure 2: Example of real world accurate starfield generated by SatSim with FOV set to 0.25 degrees. Stars are circled in red and come with a spectral type annotation which can be used for performance benchmarks for HS-ANET.
  • Figure 3: Fit rate vs. number of stars per image, comparing the original solver and HS-ANET.