热带海洋学报

• • 上一篇    下一篇

面向珊瑚礁生态系统的暗场显微原位观测技术

孙乾泰1,2,王新伟1,2,3,雷平顺1,孙亮1,刘胜4,陈嘉男1,张钰润1,周燕1,2,3,刘育梁1,2   

  1. 1. 中国科学院半导体研究所光电系统实验室,北京 100083;

    2. 中国科学院大学材料与光电研究中心,北京 100049;

    3. 中国科学院大学电子电气与通信工程学院,北京 100049;

    4. 中国科学院南海海洋研究所, 热带海洋环境与岛礁生态全国重点实验室, 广东省应用海洋生物学重点实验室, 广东 广州 510301




  • 收稿日期:2026-05-03 修回日期:2026-06-16 接受日期:2026-07-20
  • 通讯作者: 王新伟
  • 基金资助:
    国家科技基础资源调查专项(2022FY100605); 中国科学院野外站重点科技基础设施建设项目(KFJ-SW-YW047)

Dark-Field Microscopic In Situ Observation Technique for Coral Reef Ecosystems

SUN Qiantai1,2, WANG Xinwei1,2,3,LEI Pingshun1, SUN Liang1, LIU Sheng4, CHEN Jianan1, ZHANG Yurun1, ZHOU Yan1,2,3, LIU Yuliang1,2   

  1. 1. Optoelectronic System Laboratory, Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China;

    2. Center of Materials Science and Optoelectronics Engineering, University of Chinese Academy of Sciences, Beijing 100049, China;

    3. School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China;

    4. State Key Laboratory of Tropical Oceanography, Guangdong Provincial Key Laboratory of Applied Marine Biology, South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, China;




  • Received:2026-05-03 Revised:2026-06-16 Accepted:2026-07-20
  • Supported by:

    Science & Technology Fundamental Resources Investigation Program (2022FY100605); Key Research Infrastructures in CAS Field Stations (KFJ-SW-YW047)

摘要: 针对珊瑚礁生态系统中珊瑚组织、共生虫黄藻及浮游生物等微尺度目标原位观测能力不足的问题,本文提出并研制了珊瑚礁双模态暗场显微原位观测系统“金睛”,并引入人工智能AI技术建立了量化分析流程,以满足珊瑚礁底栖界面及上覆水体中20μm至1mm尺寸范围内的水下目标观测需求。该系统采用了片光与环光双模态差异化照明以适配不同观测场景:金睛-Coral采用正交片状照明,用于珊瑚底栖界面的抵近显微观测;金睛-Plankton采用侧向环形照明,用于上覆水体垂直剖面中浮游生物及悬浮颗粒的显微观测。在实验中,金睛在光衰减系数不大于25.63 m⁻¹的浑浊条件下均能保持较好的目标可见性,即使在光衰减系数高达29.88 m⁻¹的高浑浊海水环境中仍具备有效成像能力。在AI量化分析中,通过图像增强、YOLOv8目标检测与计数以及粒径统计等环节,实现生物丰度估计和粒径分布等定量分析。实验结果表明,AI量化分析流程能够稳定实现目标检测、丰度估计与粒径统计,在较高浓度藻类样品中丰度估计相对误差优于6%。本文研制的系统已用于西沙珊瑚礁生态系统调查,相关研究可为珊瑚礁生态监测及相关生态过程研究提供技术支撑。

关键词: 珊瑚礁, 显微成像, 暗场成像, 人工智能, 量化分析

Abstract: To address the limited capability for in situ observation of microscale targets in coral reef ecosystems, such as coral tissues, symbiotic zooxanthellae, and plankton, this study proposes and develops a dual-modal dark-field microscopic in situ observation system for coral reefs, named “Jinjing”. An artificial intelligence (AI)-based quantitative analysis workflow is also introduced to meet the underwater observation requirements for targets ranging from 20μm to 1mm at the coral reef benthic interface and in the overlying water column. The system adopts two differentiated illumination modes, namely light-sheet illumination and annular illumination, to accommodate different observation scenarios. Jinjing-Coral employs orthogonal sheet illumination for close-range microscopic observation of the coral benthic interface, while Jinjing-Plankton uses lateral annular illumination for microscopic observation of plankton and suspended particles in vertical profiles of the overlying water. Experimental results show that Jinjing maintains good target visibility under turbid conditions with a light attenuation coefficient of up to 25.63 m⁻¹, and remains capable of effective imaging even in highly turbid seawater with a light attenuation coefficient as high as 29.88 m⁻¹. In the AI-based quantitative analysis workflow, image enhancement, YOLOv8-based target detection and counting, and particle size statistics are integrated to enable quantitative analyses such as biomass abundance estimation and particle size distribution. The results demonstrate that the AI workflow can stably perform target detection, abundance estimation, and particle size analysis, achieving a relative error of less than 6% in abundance estimation for high-concentration algal samples. The developed system has been applied in ecological surveys of coral reef ecosystems in the Xisha Islands, and this study provides technical support for coral reef ecological monitoring and related ecological process research.

Key words: coral reef, microscopic imaging, dark-field imaging, artificial intelligence, quantitative analysis