Svenesis Satellite Trail Cleaner

Version 1.0.0 – GPL-3.0-or-later

The Satellite Trail Cleaner identifies and removes linear satellite or aircraft trails from individual astrophotography sub-exposures before stacking. Unlike Siril’s standard sigma-clipped rejection – which requires 8+ frames to work reliably – this tool cleans affected frames individually, making it invaluable for short sequences, single-night campaigns, or LRGB stacks with limited per-filter counts.

When to Use This Tool

✅ Best suited for

• Short sequences (typically fewer than 8 frames per filter) where Siril’s sigma-clipped stack rejection cannot reliably remove trails
• Single-night campaigns and LRGB stacks with limited per-filter counts
• Calibrated sub-exposures – FITS, XISF, TIFF, or RAW that has already been calibrated (darks, bias, flats)
• Linear trails: satellites (including “flashing” satellites with bright pearls), aircraft, and bright meteor streaks

⛔ When NOT to use it

• Uncalibrated RAW files with strong sensor banding – the built-in banding detector classifies frames as clean / moderate / strong / severe and will warn you. Calibrate first (darks/bias) before running the cleaner.
• Large, well-distributed stacks (8+ frames per filter) – Siril’s standard sigma-clipped rejection handles those reliably and is the better choice. The cleaner is designed to fill exactly this gap, not to replace stack-based rejection.
• Curved or non-linear features (smoke trails, fish-eye distortion) – the Radon transform expects straight lines
• As a substitute for calibration – the tool repairs trails, not gradients, hot pixels, or noise patterns

Detection – STScI Median Radon Transform

The script uses the same satellite-detection pipeline NASA’s Space Telescope Science Institute employs on HST/ACS imagery (Stark et al. 2022, ACS ISR 2022-08 – PDF). The Median Radon Transform is mathematically robust to bright stars, preventing the false-positive “fan” patterns that plague traditional Hough or Canny detectors.

• Matched-filter peak detection with three precomputed line-width kernels
• Per-candidate validation via Gaussian fitting
• Optional persistence testing to reject non-uniform features
• Automatic endpoint extension to image boundaries
• Bright-halo mask growth absorbs satellite flashes and pearls

Inpainting – Six Methods with Smart Recommendations

The tool offers six inpaint strategies, each suited to different trail characteristics:

1. Perpendicular Strip Median (default) – preserves sky gradients; excellent for “flashing satellite” trails with bright pearls
2. Harmonic / Laplace (∇²u = 0) – maximum-principle solver preventing ringing
3. Nearest Neighbor + Smooth – fast, ~1 s per frame
4. OpenCV Fast Marching (Telea) – sub-200 ms, good fallback
5. OpenCV Navier-Stokes – fastest option
6. Biharmonic (experimental) – mathematically smooth but may overshoot

Sky-noise matching: a post-processing step adds Gaussian noise matching the local sky σ, making filled regions statistically indistinguishable from real sky for stack-rejection algorithms.

Intelligent Workflow Features

• Recommendation banner – analyzes the specific trail profile (sky gradient, pearl count, mask compactness) and recommends the best-suited method with a one-click Apply button
• Star protection (default ON) – isotropy filter distinguishes radially-symmetric stellar peaks from elongated satellite pearls, preventing accidental removal of real stars beneath trails
• Interactive line picker – every detected trail appears as a clickable green overlay; toggle remove/keep before applying
• Workflow advisor banner – color-coded guidance (red / orange / yellow / hidden) indicates whether the cleaner is appropriate for your folder

File Format Support

Format-preserving round-trips for all common astrophotography formats:

• FITS – header preserved verbatim; plate-solving data retained
• XISF (PixInsight native) – FITSKeywords and XISFProperties carried forward, including astrometric solutions
• TIFF – bit-exact dtype control; PlanarConfiguration honored
• RAW (CR2/CR3/NEF/ARW/DNG/ORF/PEF/RAF/RW2/SRW/MRW/X3F/KDC) – debayered to FITS; original preserved

Original files move to an originals/ subfolder – all changes are reversible.

Batch Processing & Performance

• Parallel pipeline – frame N+1’s load and detection runs in a worker thread while the main thread inpaints and writes frame N (~1.5–2× speedup on 20+ frame batches)
• Progress feedback – running counters for Cleaned / Skipped / Errors plus elapsed and remaining time
• Thread management – cross-platform BLAS thread sanity prevents oversubscription on macOS Apple Silicon, Linux, and Windows

Output & Documentation

Non-destructive results:

• Cleaned images replace original filenames (existing pipelines unaffected)
• Audit trail: trail_cleanup_report.txt (human-readable TSV) and .json (machine-structured records per frame)
• Frames with no detected trails remain untouched

The audit log captures detection status, line count, pixels replaced, inpaint method used, dilation settings, scan mode, RGB-reduce mode, version, and timestamp.

Dependencies

sirilpy (bundled); auto-installed: numpy, PyQt6, astropy, opencv-python-headless, photutils, scikit-image, acstools, xisf, tifffile