Chad Mello, U.S. Air Force Academy; Brandon Son, U.S. Air Force Academy; Denis Irimiea, U.S. Air Force Academy; Nicholas Wright, U.S. Air Force Academy; David Strong, Strong EO Imaging, Inc.; Timothy Giblin, i2 Strategic Services LLC; Francis Chun, U.S. Air Force Academy
https://doi.org/10.64861/TCME5405
Keywords: SDA, SSA, Falcon Telescope Network, satellite spectroscopy, slitless spectroscopy, transmission grating, spectral imagery, edge computing, automation pipeline, data ingestion, image preprocessing, spectral extraction, quality screening, metadata indexing
Abstract:
Images of low-resolution satellite spectra (103 < R < 147 over 350 nm < λ < 700 nm) collected using transmission gratings are routinely produced by telescopes operated by the Center for Space Situational Awareness Research (CSSAR) at the U.S. Air Force Academy and related nodes in the Falcon Telescope Network (FTN). While calibration and spectral extraction steps are established, operational throughput and timeliness are often limited by (i) mount- and geometry-dependent imaging conditions, and (ii) the continued need for human-in-the-loop inspection to identify contaminated frames prior to downstream exploitation. This paper presents current work and a general system architecture for an end-to-end FTN automation pipeline, including the overall flow, functional components, and deployment concept. The proposed approach is intended to (1) poll each telescope/instrument for new data products, (2) ingest and preprocess imagery at the point of collection, (3) perform automated quality screening and triage, and (4) publish resulting products and metadata into an indexed, online data store for rapid discovery and consumption. The architecture is designed around local edge compute at each telescope site to reduce latency and enable near-real-time access to quality-scored imagery and extracted spectral products. Machine learning and other AI techniques may be integrated as downstream consumers (e.g., dimensionality reduction, classification, regression), but the focus of this work is the automation and quality triage needed to make such analysis timely and scalable. A central technical element is an algorithmic and rule-based quality-screening framework that flags image quality prior to final spectral exploitation. Operating directly on preprocessed 2D spectral images, the framework: (1) detects the zeroth- and first-order signals using flux-weighted box/centroid searches to establish a 1D extraction path after background estimation/subtraction; and (2) applies interpretable, threshold-based checks to identify common contamination and failure modes, including star-streak intersection, partial first-order coverage, elevated and non-uniform backgrounds (e.g., lunar contamination), background gradients, low signal-to-noise, and overexposure/saturation. Each frame is assigned an interpretable traffic-light disposition (red/yellow/green) along with per-flag metadata to standardize triage and reduce analyst touch time. The proposed system is intended to increase processing throughput and reduce time-to-access for space-domain-awareness-relevant spectral data products across FTN via reducing manual pre-screening and enabling edge-to-database publication
Date of Conference: September 15-18, 2026
Track: Satellite Characterization