文献类型：期刊文献

中文题名：Hyperion高光谱数据森林郁闭度定量估测研究

英文题名：Estimating forest crown closure using Hyperion hyperspectral data.

作者：谭炳香[1] 李增元[1] 陈尔学[1] 庞勇[1] 雷渊才[1]

第一作者：谭炳香

通信作者：Tan, B.-X.

机构：[1]中国林业科学研究院资源信息研究所

年份：2006

卷号：28

期号：3

起止页码：95-101

中文期刊名：北京林业大学学报

外文期刊名：Journal of Beijing Forestry University

收录：CSTPCD;;EI(收录号：2006289996629);Scopus(收录号：2-s2.0-33745752722);北大核心:【北大核心2004】；CSCD:【CSCD2011_2012】；

基金："863"国家高新技术项目(2002AA133050)

语种：中文

中文关键词：高光谱遥感;森林郁闭度;Hyperion;PCA

外文关键词：hyperspectral remote sensing, forest crown closure, Hyperion, PCA

分类号：S758.58

摘要：该文探讨和评价了利用星载EO-1 Hyperion高光谱遥感数据定量估测森林郁闭度的能力.高光谱数据光谱特征空间降维采用两种方式,一种是光谱特征选择的波段选择法(BS),另一种是光谱特征提取的主成分变换法(PCA).从森林资源变化图上获取200个样点的实测郁闭度值,130个用于建模,70个用于验证.对应图像的取值采用单像元值(NP)和3×3窗口的平均值(W33)两种方法.两种光谱特征降维方式和两种图像取值方法构成4种估测模型(BS-NP、BS--W33、PCA-NP和PCA-W33).分析过程为:①对图像进行预处理,选出质量好的波段;②采用逐步回归技术提取与郁闭度相关性高的波段或变量;③建立多元回归模型估测郁闭度;④估测精度验证.经检验,估测精度分别为:BS-NP 83.17%、BS-W33 84.21%、PCA-NP 85.62%和PCA-W33 86.34%.初步结果表明,光谱特征提取的主成分变换分析法比光谱特征选择的波段选择法的郁闭度估测更有效;3×3窗口的图像取值方法比单像元取值方法的估测精度高.
In this paper, the capability of EO-1 Hyperion data in the estimation of forest crown closure （FCC） was assessed. A comparison of two feature extraction methods was made for estimating FCC. The methods are individual band selection （BS） and principal component analysis （ PCA）, which were used to calculate plot image signature： nearest pixel value （NP） and mean value of pixels of 3 by 3 windows （W33）. Therefore, four methods were used to estimate FCC, which were BS-NP, BS-W33, PCA-NP and PCA-W33. Hyperion data were acquired on July 14, 2001. A total of 200 FCC field sample plots were selected from the forest resources dynamic map, of which 130 for developing model and 70 for validation. The analysis procedure consists of： 1 ） preprocessing of hyperspectral data, including vertical stripes removing, smile effect reduction and atmospheric correction; 2） extracting features with the two methods： BS and PCA using stepwise regression technique; 3 ）establishing multivariate regression prediction models and predicting forest crown closure; 4） validating the FCC estimating results with the sample plot data. The accuracy for BS-NP, BS- W33, PCA-NP and PCA-W33 were 83.17%, 84.21%, 85.62% and 86.34% respectively. The primary results indicate that the features extracted with PCA method are more effective for estimating FCC than those with BS method. Results also prove that in predicting and mapping FCC, W33 method has higher precision than the nearest pixel method.