【作者】 郑海永；

【导师】 姬光荣；

【作者基本信息】 中国海洋大学 ， 海洋信息探测与处理， 2009， 博士

【摘要】 近年来，近海海域赤潮频发，造成严重危害，引起了世界各国政府、公众和科学家的关注。而我国近海赤潮问题也早已引起了公众、政府和科技界的广泛关注，国家从基础研究、高技术发展等不同层面安排研究赤潮发生的机制、预警、预报与防治方法，着手建立业务化的赤潮监测体系等。其中，快速有效的鉴定赤潮主要藻种是赤潮自动监测中的一个重要环节。本文主要针对我国沿海常见有害赤潮藻，采集藻种不同生长时期、不同角度的多视点图像，建立中国海常见有害赤潮显微图像数据库；以传统的生物形态分类学为依据，利用图像分析、统计学习和模式识别技术，研究赤潮藻种的多视点综合表示特征，建立赤潮藻显微图像识别系统；同时，构建一个基于Web的有害赤潮显微图像采集与诊断系统，可对上传的显微光学图像提供远程处理、分析与反馈。主要工作包括：1．中国海常见赤潮藻的生物形态学特征和分类学研究。结合中国沿海近年来的赤潮发生情况，列出了中国海常见赤潮藻种名录(共40种，其中包括11种重点藻种)；通过光学显微镜人工目视和拍照观察等手段，研究了这40种赤潮藻的生物形态学特征；并在此基础上，初步设计了这40种赤潮藻的形态学分类系统，从而为有害赤潮显微图像数据库系统和诊断系统的设计提供了依据，同时为下一步显微图像识别系统的研究奠定了基础。2．有害赤潮显微图像数据库系统和诊断系统的设计与实现。以40种所研究赤潮藻的生物形态学分类特征为依据，结合显微图像识别系统的实际应用需求，设计并搭建了有害赤潮显微图像数据库系统；结合JAVA(J2EE)技术和Web数据库技术，设计并实现了基于Web的有害赤潮显微图像采集与诊断系统。截止目前，该有害赤潮显微图像数据库已经收录课题所研究40种中国海常见赤潮藻中的24种(其中包括11种重点藻种中的10种)，以甲藻为主，包括甲藻门15种、硅藻门4种、硅鞭藻纲3种、定鞭藻纲1种、蓝藻门1种，显微图像共计3790张；并且，所建立的有害赤潮显微图像数据库系统和基于Web的有害赤潮显微图像诊断系统已经分别正常运行一年和半年多。3．显微图像识别系统研究。基于中国海常见赤潮藻的生物形态学分类知识，以已经设计实现并运行的有害赤潮显微图像诊断系统为平台，提出了显微图像识别系统的技术路线，并针对显微图像识别进行了如下工作：●提出了一种基于多方向投影的自动阈值方法进行藻种显微图像分割，该方法能较好的用于无角毛类藻种显微图像的目标细胞分割，不但能够提取藻种显微图像中的主要目标细胞，在此主要目的基础上尽可能大的保留原图像中目标细胞的全部信息，而且由于采用了较多方向(八个方向)进行投影，因此能够更为精确的定位细胞的位置，从而完全去除周围的杂质。该工作为下一步藻种图像细胞特征提取和分类识别奠定了基础。●从理论上分析了变分偏微分方程图像分解的TV-L~2模型和TV-G模型及其数值解法，结合这两种模型在藻种显微图像分解中的实际应用，进一步验证了TV-G模型中G空间对纹理的描述优于TV-L~2模型中的L~2空间，从而能够较为精确的实现细胞图像纹理与主体结构的分解；分别采用具有尺度、平移、旋转不变性的Hu不变矩来描述藻种显微图像分解后的主体结构(形状)特征，具有多尺度多分辨率变化不变性的分形维数来描述藻种显微图像分解后的纹理(壳面花纹)特征，进行分类识别，并通过实验验证了该方法的可行性。●改进了基于分形邻距的图像匹配方法，针对藻种显微图像，改进方法较传统方法能够更为准确的实现目标匹配，从而可较大程度提高藻种的识别准确率。更多还原

【Abstract】 In recent years, red tides with serious harm appear more and more frequently incoastal waters, which caused more attention by the governments, the public and scientists.The same phenomenon happens in China, the country arranged researches on formationmechanism, early warning, forecasting, preventing and controlling methods ofred tide on different levels such as the fundamental research and high-technology development,which aims to establish operational monitoring system of red tide. Therefore,identifying the dominant species of red tide rapidly and effectively plays an importantrole in automatic monitoring of red tide. According to the situation that Harmful AlgalBlooms (HAB) appear in China’s coastal waters, this paper elaborates the work includingmicroscopic image acquisition of HAB on multi-viewpoints with different growingperiods and establishment of microscopic image database about HAB appearingfrequently in China’s coastal waters. Based on the traditional biological morphologicaltaxonomy, this paper studies multi-viewpoints characteristics of HAB microscopicimages, establishes the microscopic image identification system of HAB, and buildsan image collecting and diagnosing system for HAB based on B/S (Browser/Server)architecture which can provide remote processing, analysing as well as feedback foruploaded microscopic images. The work mainly contains:1. Research on ecological characteristics and taxonomy of HAB appearing frequentlyin China’s coastal waters. According to the information of red tide appearingin recent years, listing the species of red tide appearing frequently inChina’s coastal waters including 40 species totally within 11 important species;studying the ecological characteristics of these 40 species by the observation aswell as photographs through microscopes; designing the taxonomic system forthese 40 species. These work lay the foundation for designing microscopic imagedatabase of HAB and microscopic image identification system.2. Design and implementation of HAB microscopic image database and diagnosis system. Based on the biological ecological taxonomic characteristics of the 40species and the application requirements of the microscopic image identificationsystem, designing and building the microscopic image database of HAB; designingand implementing the microscopic image diagnosis system for HAB basedon B/S architecture using the technology of JAVA (J2EE) and WEB database.Until now, this database collects 24 of 40 species within 10 of 11 importantspecies, including 15 species of Dinophyta, 4 species of Bacillariophyta,3 species of Dictyochophyceae, 1 species of Haptophyceae, and 1 speciesCyanophyta. Furthermore, the image database and diagnosis system have alreadyworked more than one and half a year respectively.3. Research on microsopic image identification. Based on the ecological taxonomiccharacteristics of the 40 species and the microscopic image diagnosis system forHAB, designing the technology roadmap of microscopic image identificationsystem. About this system, this paper mainly contains the work as follows:(a) proposing a method of image segmentation for cell image based on multipledirections projection and automatic thresholding. This method can separatethe object cell from the microcopic image which is useful for cell extractionof species without setae. Therefore, object cell information in microscopicimages can be extracted. Besides, the location of the cell in microscopicimages can be located accurately to remove noises around the cell becausethis method adopts 8 directions projection. This method lays the foundationfor further features extraction and identification.(b) analysing image decomposition model of TV-L~2 and TV-G as well astheir numerical solution based on variational partial differential equationstheoretically. The actual application of these two models in decomposingphytoplankon microscopic images illustrates that the texture of image isbetter described by G space in TV-G model than that by L~2 space in TV-L~2model. Therefore, TV-G model is used to decompose microscopic imagesinto cartoon and texture components. To identify the species, the cartoonimage expressing shape of the cell is described by Hu invariant moment which have the invariance of scale, translation and rotation while the texttareimage expressing valve stria is descibed by fractal dimension which havethe invariance of multi-scale and multi-resolution. And the experimentalresults prove feasibility of this method.(c) improving the image matching method based on fractal neighbor distance.The improved method can realise the object matching more exactly thantraditional method for phytoplankton images, which can improve the identificationrate dramatically.更多还原