文献类型：期刊文献

中文题名：基于机载PHI高光谱数据的森林优势树种分类研究

英文题名：Classification of forest species using airborne PHI hyperspectral data

作者：樊雪[1] 刘清旺[1] 谭炳香[1]

第一作者：樊雪

机构：[1]中国林业科学研究院资源信息研究所

年份：2017

卷号：29

期号：2

起止页码：110-116

中文期刊名：国土资源遥感

外文期刊名：Remote Sensing for Land & Resources

收录：CSTPCD;;北大核心:【北大核心2014】；CSCD:【CSCD2017_2018】；

基金：高分辨率对地观测系统重大专项项目(编号:30-Y20A37-9003-15/17);国家自然科学基金青年科学基金项目"机载激光雷达探测森林冠层高度的机理模型研究"(编号:41201334);国家高技术研究发展计划(863计划)子课题"全球林业定量遥感专题产品生产体系(二)"(编号:2013AA12A302)共同资助

语种：中文

中文关键词：高光谱数据,;PHI,;降维,;波段选择法,;SVM

外文关键词：hyperspectral data PHI dimensionality reduction band selection method SVM

分类号：TP79;TP751.1

摘要：近年来,高光谱遥感在林业方面的应用越来越广泛,尤其在分类方面居多。但机载PHI高光谱数据通常用于农业病虫害监测、海洋悬浮物颗粒监测等,在林业方面的应用较少。以湖北省荆门市东宝区为研究区,以机载PHI高光谱遥感数据为数据源,对森林优势树种进行了分类研究。首先采用独立成分分析法(independent component analysis,ICA)对裁剪后的PHI数据进行降噪,并利用自适应波段选择法(adaptive band selection,ABS)进行降维,再采用归一化植被指数(normalized difference vegetation index,NDVI)区分林地与非林地,最后利用支持向量机法(support vector machine,SVM)进行森林优势树种监督分类。研究结果表明,分类精度可达80.70%,Kappa系数达到0.75;分块处理PHI数据以及采用NDVI区分林地与非林地,对于减弱"同物异谱"和"异物同谱"现象有较好的作用;ABS与SVM相结合的分类方法,较适用于PHI数据在树种识别方面的应用探索,具有重要意义。
Hyperspectral data are becoming more and more widely used in forestry, especially in terms of classification. Nevertheless, the application of PHI in forestry is much less than that in such fields as agricultural pest and disease monitoring and marine suspended particles monitoring. PHI is used in this paper, and the study area is Jingmen in Hubei Province. This paper proposes an independent component analysis （ICA） combined with adaptive band selection （ABS） algorithm to reduce dimensions, extract forest land and non-forest land using （normalized difference vegetation index,NDVI） based on the subset images, and finally classify the images by support vector machine （SVM）, with the overall classification accuracy being 80.70%, and Kappa coefficient reaching 0.75. The results show that the chunk of PHI data and the use of the extraction of NDVI to distinguish between forest land and non-forest land to decrease the effect of “the same object with different spectra” and “the same spectrum with different objects” can yield a good effect. It is shown that the combination of ICA - ABS and SVM is suitable for PHI data. This study has an important significance for the application of hyperspectral in tree species recognition.