高光谱遥感影像森林信息提取方法比较  

Comparison among methods that extract forest information from hyper-spectral remote sensing image

作　　者：张雨[1] 林辉[1] 臧卓[1] 严恩萍[1] 东启亮[1] 

机构地区：[1]中南林业科技大学林业遥感信息工程研究中心,湖南长沙410004

出　　处：《中南林业科技大学学报》2013年第1期75-79,共5页Journal of Central South University of Forestry & Technology

基　　金："十二五"国家高技术研究发展计划(863计划)课题(2012AA102001):"数字化森林资源监测关键技术研究";林业公益性行业科研专项(201104028):"林分结构与生长模拟技术研究";国家重大专项项目(E0305/1112/02);湖南省高校科技成果产业化培育项目(11CY019)

摘　　要：以湖南省株洲市攸县黄丰桥林场为研究对象,运用最小距离、马氏距离、最大似然、光谱角制图、光谱信息散度、神经网络、支持向量机7种分类方法对Hyperion高光谱数据进行森林信息提取。通过对信息提取结果进行比较分析得出:针对Hyperion影像信息提取的7种方法中,支持向量机、神经网络和马氏距离3种分类方法较适合森林信息的提取,总体精度分别为67.39%、66.30%、62.68%;最大似然法在针叶林信息提取中的效果较好,其精度为67.31%;支持向量机法最适合提取阔叶林信息,其精度为80.19%,远远高于其它6种分类方法;运用马氏距离和神经网络法提取竹林信息精度最高,分别达到了84.21%和81.58%。Taking Huangfengqiao forest farm as the studied object, which is located in Youxian county, Zhuzhou City, Hunan Province, the forest information was extracted from hyper-spectral remote sensing image by using 7 kinds of classification method such as minimum distance,Mahalanobis distance, maximum likelihood,spectral angle mapping, spectral information divergence, neural network, support vector machine. Comparing the results of information extraction shows that among 7 kinds of methods of Hyperion image information extraction, support vector machine, neural network and mahalanobis distance were more suitable for forest information extraction,with overall accuracy of 67.39%,66.30%,62.68% respectively. The effect of maximum likelihood method extracted the coniferous forest information was batter than other＇s with an accuracy of 67.31%. The support vector machine was most suitable for extracting the information of broad-leaved forest, with an accuracy of 80.19%, the accuracy was much higher than the other six. Mahalanobis distance and neural network mthods were the two maximam ways to extract the bamboo forest information,with accuracy of 84.21% and 81.58% respectively.

关 键 词：高光谱遥感 HYPERION数据 森林信息提取 黄丰桥林场 

分 类 号：S771.8[农业科学—森林工程]