基于高分一号卫星影像的珊瑚岛礁分类方法

Coral Reefs Classification Methods Based on GF-1 Satellite Image

在国内外学者对珊瑚岛礁分类体系的研究的基础上,提出了一种适用范围较广的分类体系,利用高分一号多光谱卫星数据,对西沙群岛华光礁和盘石屿进行了最小距离、马氏距离、最大似然、神经网络、支持向量机等基于像元的监督分类和使用面向对象分类,并对分类结果进行精度评估。两次实验结果表明,支持向量机和面向对象分类方法精度较高,且面向对象分类方法具有更好的目视效果,在一定程度上能满足当前珊瑚岛礁信息提取的需要。

According to the studies on coral reefs classification systems of domestic and overseas scholars,this article proposes a detailed classification system.Additionally,based on GF-1 multispectral image,this article performs five traditional pixel-based supervised classification methods,namely minimum distance,Mahalanobis distance,maximum likelihood,neural net classification and support vector machine classification and one kind of object-oriented classification methods in the area of Discovery Reef and Passu Keah,Paracel Islands.Then accuracy assessment is carried out.The result of two experiments presents that support vector machine classification and the object-oriented classification method can achieve higher accuracy and the latter performs better visual effect and thus can meet the information extraction requirements of coral reefs to some extent.