基于WorldView-2数据和支持向量机的红树林群落分类研究  

Mangrove Community Classification Based on World View-2 Image and SVM Method

作　　者：唐焕丽[1,2] 刘凯[1] 朱远辉[1] 王树功[3] 柳林[1,4] 宋莎[1] 

机构地区：[1]中山大学地理科学与规划学院综合地理信息研究中心∥广东省城市化与地理模拟重点实验室,广东广州510275 [2]广东省海洋发展规划研究中心,广东广州510220 [3]中山大学地球科学与地质工程学院∥广东省地质过程与矿产资源探查重点实验室,广东广州510275 [4]辛辛那提大学地理系,美国俄亥俄州辛辛那提45221

出　　处：《中山大学学报（自然科学版）》2015年第4期102-111,共10页Acta Scientiarum Naturalium Universitatis Sunyatseni

基　　金：国家自然科学基金资助项目(41001291);中山大学高校基本科研业务费专项资金资助项目(13lgpy61);教育部重点实验室系统基金资助项目(2014BGERLXT13)

摘　　要：利用遥感技术对红树林进行群落级识别在红树林的资源详查、利用和保护方面具有重要意义。基于World View-2卫星影像的光谱特征、植被指数及纹理特征信息,结合实地调查中红树林植物的生长区位信息,采用面向对象结合支持向量机(Support Vector Machine,SVM)的方法对珠海淇澳岛红树林自然保护区大围湾片区的红树林植物进行群落分类,对比分析单一尺度和多尺度两种方式的分类效果。结果表明,尽管红树林群落之间光谱反射特征相似度较高,但拥有8个光谱波段的World View-2数据在此分类中仍具有很好的应用潜力;多尺度分类结果总体精度达到84.2%(kappa系数0.794),高于单一尺度分类结果的69.8%(Kappa系数为0.616)。Using remote sensing technology in Mangrove Community Classification is very significant for surveying,taking advantage of and protecting Mangrove resource. In this study,based on the spectrum characteristics of mangroves,vegetation index and texture information calculated from World View-2 satellite imagery,we used object-oriented classification method,SVM（ Support Vector Machine）,in conjunction with field surveys to map mangrove forest at communities＇ level in Daweiwan District,Qi＇ao Island,Zhuhai. The single-scale and multi-scale classification were also compared. The results indicated that World View-2 data,a very high-resolution satellite remote sensing imagery with 8 bands are very suitable for mangrove forest classification using object-oriented and SVM method. The overall accuracy and Kappa indices for mangrove forest classification at the species level in the study area were 84. 2% and0. 794 for multi-scaled analysis and 69. 8% and 0. 616 for the single-scaled.

关 键 词：红树林群落 面向对象 支持向量机 多尺度分类 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]