期刊文献+

基于机载激光雷达和高光谱数据的树种识别方法 预览 被引量:1

Automatic identification of tree species based on airborne LiDAR and hyperspectral data
在线阅读 免费下载
分享 导出
摘要 训练样本的选取是影响监督分类精度的直接原因之一,数据空间分辨率越高,训练样本要求越准确,而人机交互训练样本选取推广力有限。利用机载高光谱(AISA)和激光雷达(LiDAR)主被动遥感数据,探讨基于高分辨率影像的训练样本自动提取技术以及适合树种识别的遥感变量。根据树木的结构和高度差异,开展树高分层掩膜试验,并计算光谱间夹角,在每个高度层中自动化优选树种的高纯度训练样本。计算植被指数、主成分分析等特征变量,基于支持向量机分类器对研究区进行树种精细分类。实验表明:通过对阔叶林、马尾松Pinus massoniana,毛竹Phyllostachys edulis,杉木Cunninghamia lanceolata,油茶Camellia oleifera的训练样本分层自动提取后再进行分类,激光雷达和不敏感色素指数变量能有效提高树种分类精度。其中高光谱+激光雷达+结构不敏感色素指数变量组合的分类精度最高,其总体精度和Kappa系数分别为89.12%和0.86,阔叶林、马尾松、毛竹、杉木、油茶的用户精度分别为75.00%,100.00%,86.36%,90.91%和96.55%。该方法对本研究区森林树种的识别是有效的。 Selection of training samples,a direct factor affecting the accuracy of supervised classification,with a higher spatial resolution image,requires more accurate training samples,but the human-computer interaction capabilities in the selection of training samples is limited.Therefore,in this study,an algorithm was provided for automatic extraction of training samples.Airborne hyperspectral data and LiDAR data were used in Gutian Mountain National Nature Reserve.The hyperspectral data were used to extract training samples automatically and variables of tree species were calculated.According to differences in structure and height of individual trees provided by the canopy height model of LiDAR,a tree height mask was made to help circumvent the problem of different objects with the same spectra and identical objects with different spectra,as far as possible.Then,the spectral angle between each pixel and training sample pixel was calculated and highly pure pixels at different heights were selected automatically.In addition,a vegetation index and principal component analysis were calculated.The precise classification of tree species was carried out by a support vector machine classifier in the study area.The experiment used a method of stratified-auto sample selection to extract the training samples of broadleaf,Masson pine,Moso bamboo,Chinese fir,and tea-oil tree forests,and then classified these five tree species.Results showed that the combination of hyperspectral data,LiDAR data,and the structure of the insensitive pigment index revealed an overall accuracy of 89.12%and a Kappa coefficient of 0.86.Using a combination of the best variables,the user accuracy was as follows:broadleaf forest--75.00%,Masson pine--100.00%,Moso bamboo--86.36%,Chinese fir--90.91%,and tea-oil tree--96.55%.Therefore,integration of different remote sensing data,stratified-auto sample selection,and hyperspectral variable selection using LiDAR and the structure insensitive pigment index were effective ways for improving tree species classificati
出处 《浙江农林大学学报》 CSCD 北大核心 2018年第2期314-323,共10页 Journal of Zhejiang Forestry College
基金 国家级大学生创新创业训练计划项目(201610341013) 国家自然科学基金资助项目(41201365) 浙江农林大学科研发展基金资助项目(2014FR004)
关键词 森林测计学 高光谱 激光雷达 分层训练样本自动提取 树种识别 光谱角填图 支持向量机 forest measuration hyperspectral LiDAR stratified-auto samples selection tree species identification spectral angle mapping support vector machine
作者简介 陶江玥,从事林业遥感研究。E-mail:954972267@qq.com。;通信作者:刘丽娟,讲师,博士,从事林业遥感研究。E-mail:llj7885@163.com。
  • 相关文献

参考文献7

二级参考文献57

共引文献90

同被引文献10

引证文献1

投稿分析

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部 意见反馈