Efficient detection of faint geostationary and cislunar objects in Rubin LSST imagery

Abstract

The Vera C. Rubin Observatory’s large aperture, wide field of view, and highly sensitive camera enable the ambitious science goals of the Legacy Survey of Space and Time (LSST). Those same features also make it possible for faint objects in geostationary or cislunar orbits, potentially near the sensitivity limit of the system, to be captured during an exposure as they streak across the field of view. Science data products are susceptible to contamination by the flux from these objects if they are not detected and mitigated, particularly for the case of low surface brightness galaxies. The apparent motion from geostationary and higher objects during the ~30 s exposure time causes streaks to span up to thousands of pixels across the CCDs. Objects that would be easily detectable if they had no apparent motion are dwarfed by the per-pixel sky noise, limiting the effectiveness of conventional streak detection techniques like Canny edge detection, Hough transforms or neural networks. While techniques like matched filtering can substantially recover signal across the streak length, performing convolutions across all possible streak angles, lengths, and locations within the 3.2 gigapixels produced by Rubin about every 40 s is computationally prohibitive. We present an efficient real-time streak detection approach using a fast Radon transform to identify candidate streak lines within the image, followed by a matched filter approach within that smaller search space to optimally recover streak endpoints. A set of preprocessing steps tailored to the on-sky data reduces the occurrence of false positives. Through this multistep process, we are able to recover detection sensitivities approaching the theoretical limit. This method is demonstrated using LSST Commissioning Camera images from the Rubin Data Preview 1, applied to the detection of synthetic streaks injected into real images.