数据与计算发展前沿 ›› 2025, Vol. 7 ›› Issue (5): 88-101.

CSTR: 32002.14.jfdc.CN10-1649/TP.2025.05.007

doi: 10.11871/jfdc.issn.2096-742X.2025.05.007

• 专刊：国产算力新力量，助力科学计算应用新发展 • 上一篇    下一篇

面向新一代望远镜的多次曝光图像叠加算法研究和国产化异构计算移植

王蕾1,*(),陕欢源2,3,4,聂麟2,颜召军2,曲涵1,郑文雯1,李国亮1   

  1. 1.中国科学院紫金山天文台，南极与射电天文研究部，江苏 南京 210023
    2.中国科学院上海天文台，上海 200030
    3.中国科学院，射电天文与技术重点实验室，北京 100101
    4.中国科学院大学，北京 100049
  • 收稿日期:2025-02-28 出版日期:2025-10-20 发布日期:2025-10-23
  • 通讯作者: 王蕾
  • 作者简介:王蕾，中国科学院紫金山天文台，高级工程师，主要研究方向为天文望远镜图像处理，引力透镜。
    本文承担工作为：算法研究和开发，数值仿真，观测图像处理，算法移植。
    WANG Lei is a senior engineer at the Purple Mountain Observatory, Chinese Academy of Sciences. His primary research interests include astronomical telescope image processing and gravitational lensing.
    In this paper, he is primarily responsible for algorithm research and development, numerical simulations, observational image processing, and algorithm porting.
    E-mail: leiwang@pmo.ac.cn
  • 基金资助:
    国家自然科学基金(12573115);国家自然科学基金(12533008);国家自然科学基金(11273061);国家自然科学基金(11825303);国家自然科学基金(11673065);载人航天工程(CMS-CSST-2021-A07);载人航天工程(CMS-CSST-2021-B01);光合基金(202302017475);光合基金(202407017555)

Research on Multi-Exposure Image Stacking Algorithms for Next-Generation Telescopes and its Porting to Domestic Heterogeneous Computing Platforms

WANG Lei1,*(),SHAN Huanyuan2,3,4,NIE Lin2,YAN Zhaojun2,QU Han1,ZHENG Wenwen1,LI Guoliang1   

  1. 1. Purple Mountain Observatory & Key Laboratory of Radio Astronomy, Chinese Academy of Sciences, Nanjing, Jiangsu 210023, China
    2. Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, China
    3. Key Laboratory of Radio Astronomy and Technology, Chinese Academy of Sciences, Beijing 100101, China
    4. University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2025-02-28 Online:2025-10-20 Published:2025-10-23
  • Contact: WANG Lei

RichHTML

0

PDF

30

摘要：

【目的】 针对新一代大视场望远镜面临的空变点扩散函数（PSF）反卷积和海量数据处理难题，提出一种高效的多次曝光图像叠加算法，并实现国产化异构计算移植。【方法】 基于HybPSF空变PSF建模方法，结合升采样与PSF反卷积联合叠加算法（UPDC），解决空变PSF的空间非线性变化问题，通过国产异构计算平台对算法核心模块进行并行化移植与优化。【结果】 仿真实验表明，新算法迭代60次后，流量守恒性较传统分块反卷积方法提升约52.3倍，图像分辨率提高至原始数据的2倍；应用于JWST观测数据时，星系形态细节信噪比显著增强。异构移植后，单卡加速比达CPU单核的160倍，处理135幅图像仅需58秒。【结论】 新算法有效解决了空变PSF反卷积与海量数据算力瓶颈，显著提升了望远镜数据质量，为CSST、LSST等大视场望远镜提供了可行的数据处理方案，并支持国产化异构计算的自主可控需求。

关键词: 多次曝光图像叠加, 空变PSF反卷积, 异构计算, 国产化移植, 计算成像

Abstract:

[Purpose] To address the challenges of spatially varying Point Spread Function (PSF) deconvolution and massive data processing for next-generation wide-field telescopes, this study proposes an efficient multi-exposure image stacking algorithm and implements its porting to domestic heterogeneous computing platforms. [Methods] A hybrid PSF modeling method (HybPSF) combined with the Upsampling and PSF Deconvolution Coaddition (UPDC) algorithm is developed to handle nonlinear spatial variations of PSF. Key modules are parallelized and optimized on the domestic accelerator (Z100L/K100). [Results] Simulation tests show that the new algorithm improves flux conservation by over 5230% compared to the traditional block-based deconvolution after 60 iterations, achieving a 2× resolution enhancement. Applied to JWST observational data, the algorithm significantly enhances the signal-to-noise ratio and morphological details of galaxies. After its porting, the single-card acceleration ratio reaches 160× compared to a CPU core, processing 135 images in 58 seconds. [Conclusions] The proposed algorithm effectively resolves the technical bottlenecks of PSF deconvolution and computational efficiency, providing a robust solution for large-field telescopes (e.g., CSST, LSST). The successful porting to domestic heterogeneous computing platforms supports China’s strategic goal of self-controlled technologies in high-performance computing.

Key words: multi-exposure image stacking, spatially varying PSF deconvolution, heterogeneous computing, domestic transplantation, computational imaging