Enhancing astrometric registration of Chinese historical Astronomical Digital Plates with deep learning

Quanfeng Xu (徐权峰), Zhengjun Shang (商正君), Shiyin Shen (沈世银)*, Yong Yu (于涌)*, Meiting Yang (杨美婷)*, Hao Luo (罗浩), Jing Yang (杨静), Zhenghong Tang (唐正宏), Jianhai Zhao (赵建海)
Shanghai Astronomical Observatory, Chinese Academy of Sciences
University of Chinese Academy of Sciences
Research in Astronomy and Astrophysics 2026

*Corresponding author

Abstract

China has systematically collected nighttime astronomical plates since 1900, creating a large historical dataset that has been digitized with optical scanners. For astrometric registration of these digitized plates, sources were first extracted using SExtractor, and then matched astrometrically with Astrometry.net and the Gaia catalog. However, suboptimal early storage conditions and subsequent environmental deterioration have impeded accurate source matching, resulting in processing failures for several thousand digitized plates. In this work, we introduce a Transformer-based classification model that takes cutouts of SExtractor-detected sources as input and leverages multi-scale feature fusion to identify trustworthy stellar sources on the plates. Trained on plates with successful astrometric calibration, our AI-based classifier was then applied to SExtractor detected sources of 1883 digitized plates, enabling us to complete the astrometric registration for 1353 of them. This AI-augmented pipeline streamlines the processing of historical plate archives and enhances their scientific value for long-term time-domain astronomical studies.

Background & Motivation

Astronomical plates are some of the most significant observational tools of twentieth-century astronomy, offering long-term sky surveys that modern surveys cannot match. Since 1900, China has systematically gathered nighttime astronomical plates, building a substantial historical dataset. Although these fragile materials have been digitized with high accuracy, they need to be astrometrically registered by cross-matching detected stellar sources with modern reference catalogs such as Gaia to be scientifically useful.

Astrometric registration workflow

The astrometric registration of the digitized plates consists of three main steps: source extraction, stellar source classification, and astrometric matching.

traditional method workflow

As illustrated below, the matching process extracts bright sources to form triangular patterns, acting as geometric 'ciphers' to be cross-matched against modern star catalogs. Finding an exact match locks in the plate's precise coordinates. To ensure the success of this algorithm, our work focuses entirely on the vital preceding step: accurately classifying and distinguishing genuine stellar sources from plate artifacts.

Status
底片处理现状
Plate Count
数量
Calibrated by Traditional Methods
已校准(由传统方法完成)
15,696
Failed by Traditional Methods
待校准(传统方法失效,目前待解决)
1,883
Raw Plate
DESI
原始退化底片 AI配准与提取结果

Astronomical Digital Plate Image Calibration

Left (Raw Plate): The original historical digital plate. Right (DESI): Precisely astrometric registration for modern sky surveys.

Challenge

However, time has left its mark on these valuable records. Suboptimal early storage conditions and subsequent environmental deterioration have left many historical plates suffering from scratches, mold, emulsion degradation, and physical fractures. These defects severely complicate the reliable detection of genuine stellar sources, causing traditional computer software to fail during the astrometric matching process and resulting in thousands of uncalibrated plates.

Our work

In our work, we introduce a deep-learning-assisted framework to salvage heavily degraded astronomical plates. We developed a Transformer-based classification model that effectively learns to distinguish trustworthy stellar sources from spurious artifacts caused by dust or scratches. By training a base model and employing metadata-guided fine-tuning to adapt to specific observational conditions (such as different telescopes and exposure times), our AI-driven approach bypasses the limitations of traditional rules.

Swin Transformer Architecture

Quantitative Results

Our AI-augmented pipeline was applied to 1,883 digitized historical plates that had previously failed astrometric registration using traditional methods. The results demonstrate the exceptional resilience of our deep learning framework against environmental deterioration.

Through the combined use of our Base Model and metadata-guided Fine-Tuned Models, we successfully completed the astrometric registration for 1,353 of these challenging plates. Remarkably, the model maintained robust feature extraction even on severely degraded data, achieving success rates of nearly 68% for plates with visible mold, scratches, or minor detachment (Grade 2 and Grade 3 plates). The model demonstrates strong resilience against moderate physical defects.

1,883
Uncalibrated Plates
(Failed by Traditional Methods)
1,353
Successfully Registered
by Our Framework
~68%
Success Rate on
Severely Degraded Data

BibTeX

@article{xu2026,
      title={Enhancing astrometric registration of Chinese historical Astronomical Digital Plates with deep learning}, 
      author={Quanfeng Xu and Zhengjun Shang and Shiyin Shen and Yong Yu and Meiting Yang and Hao Luo and Zhenghong Tang and Jing Yang and Jianhai Zhao},
      year={2026},
      eprint={2604.04714},
      archivePrefix={arXiv},
      primaryClass={astro-ph.IM},
      url={https://arxiv.org/abs/2604.04714}, 
}