Improved Identification of Satellite Trails in ACS/WFC Imaging Using a Modified Radon Transform

We present a new approach to identify satellite trails (or other linear artifacts) in ACS/WFC imaging data using a modified Radon Transform. We demonstrate that this approach is sensitive to features with mean brightness significantly below the background noise level, and it is resistant to the influence of bright astronomical sources (e.g., stars, galaxies) in most cases. Comparing with a set of satellite trails identified by eye, we find a trail recovery rate of 85% and a false detection rate (after removing diffraction spikes that are easily filtered) of 2.5%. By performing an analysis using a much larger ACS/WFC data set where false trails are identified by their persistence across multiple images of the same field, we identify the Radon Transform parameter space and image properties where our algorithm is unreliable, and estimate a false detection rate of $\sim10%$ elsewhere. We apply our method to ACS/WFC data taken between 2002 and 2022 to determine both the frequency of satellite trail contamination in science data and also the typical trail brightness as a function of time. We find the rate of satellite trail contamination has increased by approximately a factor of two in the last two decades, but there is no clear systematic evolution in the typical trail brightness. Our satellite trail identification program is available as part of the \texttt{acstools} package.

💡 Research Summary

The paper introduces a novel, fully automated method for detecting satellite trails and other linear artifacts in Hubble Space Telescope ACS/WFC images. Traditional approaches—visual inspection, elongated‑source detection, machine‑learning classifiers, Hough transforms, and the standard Radon Transform (RT)—all suffer when bright point sources (stars, galaxies) are present, because the RT sums pixel values along every possible line and a single bright source can dominate the transform, masking faint linear features.
To overcome this, the authors replace the summation with a median calculation, defining a Median Radon Transform (MRT). By taking the median of pixel values along each candidate line, the transform becomes robust against isolated bright pixels while remaining sensitive to continuous, low‑surface‑brightness trails that span many pixels.
The processing pipeline, implemented as the Python function findsatmrt in the acstools package, proceeds as follows:

1. Load calibrated FL​C files, mask bad pixels using DQ flags, and set them to NaN. Cosmic‑ray flags are retained because they sometimes overlap trails.
2. Subtract a global median background and optionally re‑bin the image 2 × 2 to speed up computation (a factor of ~10).
3. Compute the MRT using a modified version of skimage.transform.radon; for each (θ, ρ) line the median is calculated, and a length map L records how many valid pixels contributed. Lines with L < 25 are discarded as unreliable.
4. Estimate an uncertainty map σ_MRT = 1.25 σ_im / √L (σ_im is a robust image standard deviation). Identify regions with S/N > 5 using photutils.StarFinder and three Gaussian kernels (widths 15, 7, 3 px) to accommodate different trail thicknesses.
5. For each candidate, extract a cutout, rotate it so the trail runs horizontally, collapse rows to a 1‑D profile, fit a third‑order polynomial to the background (excluding the central region), then fit a Gaussian to the residual to refine width, peak intensity, and S/N. Keep candidates that exceed the S/N threshold and have widths ≤ 75 px.
6. Assess “persistence” by splitting the trail into N sections (minimum 100 px each) such that each section would have S/N ≥ 3 if the total S/N is known (N ≈ (S_total/3)²). Fit each section; sections meeting the criteria add to a persistence score, which is normalized by N. Candidates with a negative persistence score are rejected.
   The authors evaluate performance on three fronts. Theoretically, the median’s standard error scales as 1/√L, so the longest possible line across a single ACS chip (L ≈ 4580 px) yields a detection limit of ~0.1 × the background noise at S/N = 5, while near the corners (L ≈ 500 px) the limit rises to ~0.3 × noise. Simulations that inject synthetic 5‑px‑wide trails into realistic ACS‑flc images (including read noise, CTE loss, flat‑field variations, dark current, bias) confirm these limits across a range of trail brightnesses (0.05–0.5 × background).
   When applied to real ACS/WFC data spanning 2002–2022 and compared against a hand‑crafted catalog from the Frontier Fields program, the algorithm recovers 85 % of the visually identified trails and produces a false‑positive rate of 2.5 % after obvious diffraction spikes are filtered out. A larger, multi‑epoch dataset allows the authors to flag spurious detections that appear only in a single exposure; after this correction the overall false‑positive rate is estimated at ~10 %.
   Limitations are acknowledged. In extremely crowded fields (e.g., dense galaxy clusters) the median can be biased upward by the high density of bright sources, reducing the MRT’s robustness. Very narrow trails (< 2 px after re‑binning) may also be missed. The authors suggest adjusting the θ/ρ sampling, minimum L, or applying a pre‑mask of bright sources in such cases.
   Finally, the authors use the catalog of detected trails to study temporal trends. Between 2002 and 2022 the fraction of ACS/WFC exposures contaminated by satellite trails roughly doubled, reflecting the growth of the low‑Earth‑orbit satellite population. However, the median trail surface brightness shows no clear secular trend, implying that newer satellites are not systematically brighter in the optical bandpass used.
   In summary, the Median Radon Transform provides a mathematically simple yet powerful way to isolate faint, extended linear features in astronomical images while suppressing point‑source contamination. Implemented in acstools as findsatmrt, it offers the community an open‑source, reproducible tool for large‑scale archival analyses of HST data and, by extension, for upcoming missions such as JWST and the Roman Space Telescope.