摘要
湿地植被的精细识别与分类是林业遥感中一个亟待解决的难点。本研究以东洞庭湖自然保护区为研究区,以苔草、芦蒿、辣蓼、杨柳和芦苇为研究对象开展高光谱数据观测。采用数据平滑、导数变换、对数变换和归一化变换等方式对高光谱数据进行预处理,再运用PCA算法分别对其进行降维运算,随后采用马氏距离、朴素贝叶斯、Knn、径向基内核支持向量机和随机森林等分类方法对降维后的数据进行分类。结果表明:1)不同预处理方式经过PCA降维后能保持自身特有的特征;2)降维后的累计方差贡献率与分类精度不存在必然联系,主成分个数能对分类精度产生影响;3)不同的分类方法对降维后的数据灵敏度不同,随机森林和径向基内核支持向量机保持较高的稳定性。
Precise identification and classification of wetland vegetation is a difficulty of forestry remote sensing. In this study, east Dongting lake natural reserve in Hunan province as the research area, development of hyperspectral data observation taking Carex tristachya, Artemisia selengensisi, Polygomum flaccidum, Salix babylonica, Phragmites australis as the research objects. The spectral data of wetland vegetation are pretreated by data smoothing, derivative transformation, logarithmic transformation and normalization transformation. Then use PCA algorithm to reduce dimension respectively. Then the classification tests of Mahalanobis distance, naive Bayesian, Knn, RBF kernel support vector machine and random forest classification methods to reduce the dimension of the data classification test are used to classify the data after dimensionality reduction. The results show that: 1) different preprocessing methods can keep their own characteristic after reducing the dimension of PCA;2) there is no inevitable connection between the cumulative variance contribution rate and the classification accuracy after reducing dimension and the number of principal components is in?uenced by the classification accuracy;3) different classification methods have different sensitivity to the dimensionality reduction data after preprocessing, and random forest and RBF kernel support vector machines maintain high stability.
作者
李世波
林辉
葛淼
LI Shibo;LIN Hui;GE Miao(Key Laboartory of Forestry Remote Sensing Based Big Data & Ecological Security for Hunan Province,Central South University of Forestry & Technology,Changsha 410004,Hunan,China;Guizhou Forestry Survey and Design Co. Ltd.,Guiyang 550001,China,Guizhou;Shanghai Ruya Information Technology Co. Ltd,Shanghai 215008,China)
出处
《中南林业科技大学学报》
CAS
CSCD
北大核心
2019年第11期36-41,共6页
Journal of Central South University of Forestry & Technology
基金
国家自然科学基金项目（31370639）
湖南省科技厅项目（2016TP1014）
关键词
高光谱
降维
分类
主成分分析
东洞庭湖
hyperspectral
dimensionality reduction
classification
principal component analysis
east Dongting lake
