摘要
本文利用主成分分析法分别对乔木树种高光谱反射率原始数据及3种预处理数据进行降维运算,再使用SVM-RBF、SVM-Linear、BP、Fisher 4种分类算法,对降维后的数据进行分类测试,发现累积方差贡献率与分类精度没有必然联系,而主成分的个数对分类结果的影响较为明显;不同的数据预处理方法和不同的分类方法对主成分分析算法降维后数据的分类灵敏度不同。
In order to investigate the separability of tree species using hyperspectral data, the three different data transformations and dimensional reduction of the hyperspectral reflectivity data using Principal Component Analysis （PCA） algorithm were explored in the paper. Four classification algorithms including Support Vector Machine （SVM）-Raial Basis Function （RBF）, Support Vector Machine （SVM）-Linear,Back Propagation（BP）neural network and Fisher classification method were compared. The results showed that cumulative contribution of variance was not necessarily associated with classification accuracy. However, the number of principal components had a more obvious effect on classification. Various data transformations and classification methods showed the different classification effects.
出处
《测绘科学》
CSCD
北大核心
2014年第2期146-149,共4页
Science of Surveying and Mapping
基金
国家自然科学基金项目(30871962)
国家自然科学基金项目(31100412)
国家"十二五"863项目(2012AA102001):数字化森林资源监测关键技术研究
国家重大专项(E0305/1112/02)
关键词
高光谱
降维
分类
主成分分析
黄丰桥林场
hyperspectrum
dimension reduction
classification
Principal Component Analysis（PCA）
Huangfengqiao forestry center
