引用本文:	孙晓霞,孙松,王世伟,刘梦坛,赵永芳.图像自动识别技术在胶州湾浮游动物生态学研究中的应用.海洋与湖沼,2011,42(5):647-653.

【打印本页】【HTML】【下载PDF全文】【查看/发表评论】【下载PDF阅读器】【关闭】

过刊浏览高级检索

本文已被：浏览 3253次下载 3374次	码上扫一扫！
分享到：微信更多字体:加大+\|默认\|缩小-
图像自动识别技术在胶州湾浮游动物生态学研究中的应用
孙晓霞¹, 孙松^1,2, 王世伟¹, 刘梦坛¹, 赵永芳¹
1.山东胶州湾海洋生态系统国家野外科学观测研究站;2.中国科学院海洋研究所海洋生态与环境科学重点实验室

摘要:

结合 Zooscan 扫描技术与 ZooProcess 分析与图像自动识别方法, 进行了胶州湾浮游动物图像自动识别的研究。通过对胶州湾 2009 年浮游动物样品进行标准化扫描, 随机选取不同类群的浮游动物图像, 建立胶州湾浮游动物图像培训数据库并进行性能验证, 表明对胶州湾绝大部分类群, 图像识别的准确率可以达到 80%以上, 且误判率低于 20%。对于毛颚类、桡足类、夜光虫、磷虾等的识别准确率可以高达 90%以上。进一步将图像自动识别结果与人工分类的结果进行比较, 发现对于胶州湾5个主要的优势类群, 两种方法之间存在极其显著的相关性, 尤其是桡足类和毛颚类, R²值分别可达到 0.96 和 0.75。在此基础上, 进一步分析该图像识别方法在胶州湾浮游动物体积变动、粒级组成中的应用, 为利用图像手段进行胶州湾浮游动物生态学及长期变化的研究奠定基础。

关键词: 浮游动物, 图像, 自动识别, 胶州湾

DOI：10.11693/hyhz201105004004

分类号:

基金项目:中国科学院知识创新工程重要方向项目群项目, KZCX2-YW-Q07-01 号; 国家“973”项目, 2011CB403603 号; 国家自然科学基金项目, 40876083 号, 40631008 号; 国家“973”项目, 2006CB400606 号; 国家海洋局公益项目, 200805042 号。

附件

APPLICATION OF AUTOMATED IMAGE IDENTIFICATION IN ZOOPLANKTON ECOLOGY STUDIES IN THE JIAOZHOU BAY

SUN Xiao-Xia¹, SUN Song^1,2, WANG Shi-Wei¹, LIU Meng-Tan¹, ZHAO Yong-Fang¹

1.Jiaozhou Bay Marine Ecosystem Research Station, Institute of Oceanology, Chinese Academy of Sciences;2.Key Laboratory of Marine Ecology and Environmental Sciences, Institute of Oceanology, Chinese Academy of Sciences

Abstract:

Zooplankton plays an important role in the marine ecosystem. How to rapidly identify zooplankton species is a key problem in zooplankton ecology studies. Automated zooplankton image identification technique is a rapid and standard method developed in recent years. However, this technique has not been used efficiently in zooplankton research in China. By combing the approaches including Zooscan, Zooprocess, and Plankton Identifier, we used the automated image identification method in the Jiaozhou Bay for the first time. A learning set of Jiaozhou Bay zooplankton images were set up according to the dominant zooplankton composition. Results of the performance test indicated that the recall was higher than 80%, and contamination was lower than 20% for most zooplankton groups. For the groups of Copepod, Chaetognath, Noctiluca, and Euphausia, the recall was higher than 90%. When comparing the results obtained from both automated and manual identification, we found that there was significant correlation between the two methods among the five dominant groups, especially for the group of Copepod and Chaetognath, and the values of R² reached 0.96 and 0.75, respectively. The automated analysis was further used for the study of biovolume and size spectra of zooplankton, which improved that the automated image identification was very useful for zooplankton ecological study and long term change research in the Jiaozhou Bay and other coastal ecosystems.

Key words: Zooplankton, Image, Automated identification, Jiaozhou Bay