Cosmic Ray Detection and Rejection for CSST Facilities: CSST Source Extractor ( E. Bertin & S. Arnouts 1996 )

Yan Yu, Bin Ma, Tianmeng Zhang, Yi Hu, Yajie Zhang · 2025

Read Cosmic Ray Detection and Rejection for CSST Facilities: CSST Source Extractor ( E. Bertin & S. Arnouts 1996 ) by Yan Yu, Bin Ma, Tianmeng Zhang, Yi Hu, Yajie Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01524.txt.

Introduction

Yan Yu and four colleagues prepare image-cleaning tools for the China Space Station Survey Telescope, whose survey images will be peppered by cosmic ray hits. The telescope revisits each patch of sky only rarely. So the usual trick of stacking several frames to spot the damage will not work. Each image must be cleaned on its own.

The team retrained the DeepCR detection model on simulated telescope images, reaching recall near 97.9 percent and precision near 98.7 percent on cosmic ray detection. A second, morphology-aware step repairs the damage: a UNet++ network restores scientifically valuable regions such as stars and galaxies, while adaptive median filtering handles plain background. Corrected targets end up with measurement errors close to clean ones, and detection rates rise about 13.6 percent over simple masking.

Catalog record

CREATORYan Yu, Bin Ma, Tianmeng Zhang, Yi Hu, Yajie Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01524

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Cosmic Ray Detection and Rejection for CSST Facilities: CSST Source Extractor ( E. Bertin & S. Arnouts 1996 ) Yan Yu, Bin Ma, Tianmeng Zhang, Yi Hu, Yajie Zhang Abstract As a space telescope, the China Space Station Survey Telescope (CSST) will face significant challenges from cosmic ray (CR) contamination. These CRs will severely degrade image quality and further influence scientific analysis. Due to the CSST’s sky survey strategy, traditional multi-frame stacking methods become invalid. The limited revisits prompted us to develop an effective single-image CR processing method for CSST. We retrained the DeepCR model based on CSST simulated images and achieved 97.90 ± 0.18 % 97.90\pm 0.18\% recall and 98.67 ± 0.05% precision on CR detection. Moreover, this paper puts forward an innovative morphology-sensitive inpainting method, which focuses more on areas with higher scientific value. We trained a UNet++ model especially on contaminated stellar/galactic areas, alongside adaptive median filtering for background regions. This method achieves effective for CRs with different intensities and different distances from centers of scientific targets. By this approach, the photometric errors of CR-corrected targets could be restricted to the level comparable to those of uncontaminated sources. Also, it increases the detection rate by 13.6% compared to CR masking. This method will provide a robust CR mitigation for next-generation space telescopes. I Introduction As an important next-generation astronomical observation facility, the China Space Station Survey Telescope (CSST) is China’s first large-scale optical astronomical telescope located in outer space ( H. Zhan 2021 ) . Operating in the unique space environment, this space-based telescope faces distinct challenges compared to ground-based ones. Without atmospheric protection, CSST will be exposed to a large amount of cosmic rays(CRs), which can severely contaminate its images. CR contamination occurs when charged particles in CRs pass through solid-state detectors, such as CCD or CMOS sensors. These particles generate electron-hole pairs, causing accumulation of excess charges in the affected pixels. CRs from different sources leave signals with various intensities on detectors. In astronomical images, CRs appear as randomly distributed over-bright pixels with distinct and sharp edges. Depending on the incident angle, CRs typically manifest as point-like features that may span several pixels, as well as elongated track-like or worm-like patterns that can extend from tens to nearly a hundred pixels ( M. Niedzwiecki et al. 2019 ) . CR contamination introduces significant noise in astronomical observations, which greatly degrades image quality and compromises subsequent scientific analyses ( B. Mirko et al. 2020 ; D. Groom 2002 ) . According to N. D. Miles et al. 2021 , a single CR event affects approximately 7 pixels on average and deposits about 2700 electrons on the Hubble Space Telescope (HST). These values are specific to the HST’s detectors and can vary significantly depending on the size and thickness of the image sensor used. This study also provides complete statistics on the total events, their morphologies, and the lengths of contamination paths across various instruments. Furthermore, the Euclid VIS instrument (with 566 s exposures) Quick Data Release (Q1) shows an average CR contamination rate of 2.05%. In some localized areas, this rate reaches as high as 5%, displaying characteristic ”X-ray” spatial patterns ( E. Collaboration et al. 2025 ) . As for the CSST, with 150 seconds exposures, simulation results indicate that approximately 200,000 pixels per image are affected by CRs, which accounts for 2–3% of the effective pixels in the astronomical observations (Fang et al., 2025, submitted). Therefore, effective mitigation of these artifacts is critical to maintaining the telescope’s image quality and the validity of research findings. Utilizing the fact that CRs rarely strike the same pixel in multiple exposures, CR removal can be easily achieved through multi-frame image stacking analysis. By comparing several (at least 3) distinct frames, contaminated pixels can be identified and replaced using median values from unaffected frames, such as AstroDrizzle did for the HST images ( R. J. Avila et al. 2014 ) . However, the CSST wide-field survey strategy provides only a single pass per detector for any given sky region. Although observations in each band (utilizing 2-4 independent detectors) yield 2-4 revisits, these exposures exhibit significant temporal discontinuities. Such sparse and non-contiguous observations make it impossible to remove CRs using multi-frame median stacking. Besides, multi-frame stacking methods face serious limitations: they become improper when observations lack consecutive exposures or when the target’s variability timescale is comparable to the exposure time. Additionally, variations in the point spread function (PSF) between exposures can lead to systematic errors in traditional multi-frame methods ( E. R. James 2000 ) . Consequently, CR correction methods that can operate on individual images must be developed and deployed. Traditional methods of single-frame CR detection could be primarily categorized into two classes: statistical and morphological. Statistical approaches include sigma-thresholding for outlier identification based on the standard deviation of pixel values. And the histogram-based method ( P. Wojtek 2003 ) locates outliers based on assumptions of a Gaussian distribution. E. R. James 2000 developed PSF-matched filtering for CR detection, because CRs are not affected by the telescope’s PSF. Morphological methods, on the other hand, include the L.A.Cosmic algorithm ( P. G. van Dokkum 2001 ) , which detects CRs by identifying sharp-edged features through Laplace edge detection and analyzing local symmetry patterns, effectively detecting CRs from under-sampled point sources. Alternatively, the fuzzy logic approach ( L. Shamir 2005 ) describes images using three key characteristics (target size, edge gradient sharpness, and overall image sharpness), and detects CRs by simulating a human’s visual inspection. Recent advances in machine learning have significantly improved CR detection accuracy and adaptability. Early studies used decision trees ( S. Steven et al. 1995 ) and artificial neural networks ( W. Waniak 2006 ) for star-CR classification. Deep learning breakthroughs include UNet based DeepCR model ( K. Zhang & J. S. Bloom 2020 ) , which enables end-to-end detection and inpainting, and on this basis Cosmic-CoNN network ( C. Xu et al. 2023 ) addresses instrument dependency and class imbalance issues through enhanced generalization capabilities. Z. Chen et al. 2024 also improved the DeepCR model by utilizing CRs in dark frames to eliminate the need for multiple frames of the same fields. Their retrained DeepCR model can detect approximately 7% more good-quality stars. These developments provide optimized solutions for CR processing in next-generation telescopes such as CSST. While current CR detection methods demonstrate reliable performance through diverse algorithmic approaches, CR inpainting remains predominantly reliant on simple median/mean filtering techniques, with DeepCR’s machine learning approach being a notable exception. These conventional methods exhibit critical limitations in repairing large-area CR contamination, and particularly, when CRs overlap scientifically vital regions such as galaxies or stars—precisely the primary targets of astronomical investigations rather than full-frame background areas. Several general image restoration techniques can also be applied to CR removal. For instance, OpenCV’s built-in inpaint function offers two modes based on the Fast Marching Method (FMM) and Navier-Stokes equations, suitable for repairing small-area discrete CR hits. Additionally, maskfill ( P. van Dokkum & I. Pasha 2024 ) provides a robust method for filling masked regions in astronomical images. Machine learning approaches include a generative adversarial network (GAN) model ( K. Schawinski et al. 2017 ) that achieves simultaneous noise suppression

Read the complete text