基于形状特征描述子的赤潮优势种识别研究  

Red Tide Dominant Species Recognition Research Using Shape Descriptors

作　　者：郑少平[1] 谢杰镇[1] 王博亮[1] 江涛[2] 

机构地区：[1]厦门大学信息科学与技术学院,福建厦门361005 [2]国防科技大学电子科学与工程学院,湖南长沙410073

出　　处：《厦门大学学报（自然科学版）》2009年第5期676-679,共4页Journal of Xiamen University：Natural Science

基　　金：福建省青年科技人才创新项目(2009J05156)资助

摘　　要：研究了赤潮优势种识别技术,该技术运用在基于流式细胞技术的赤潮实时监控系统中.由于不同生长周期和拍摄视角对藻类细胞的形态有较大的影响,仅用一种特征就将藻类分开是不现实的.采用3种形状特征描述子提取特征,再利用多级主特征向量评估准则进行特征融合和降维,最后使用SVM进行分类.通过对6类我国附近海域常见赤潮优势种的3000个样本的自动分类实验结果表明,识别精度可以达到95.80%,与人工识别精度相当,具备较高的实用性.The research is about the red tide dominant species recognition,which is used by flow-cytometry-based real-time red tide monitoring system. As the shape of the algae have a great difference at different growth cycles and visual angles,it is impossible to discriminate the red tide dominant species using one type of feature. The method proposed in this paper is： using three types of shape descriptors to extract feature,multilevel dominating eigenvector estimation criteria was used to reduce redundancy, and finally,using SVM as the classifier. The experiment for classifying 6 classes 3 000 samples of red tide dominant species,which are common in the sea off china,indicated that the recognition precision could be achieved to 95.80%. It is quite equal to artificial recognition accuracy which has the high usability.

关 键 词：形状特征描述子 自动识别 模式识别 主成份分析 赤潮优势种 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]