基于微分变换的高光谱马尾松和杉木识别

Classification of Pinus massoniana and Cunninghamia lanceolata using hyperspectral image based on differential transformation

高光谱遥感能分辨出地物间微小反射光谱差异信息,可用于解决林种遥感分类光谱识别的难题。利用Hyperion高光谱遥感影像,结合地面实测林种样地,对安徽省黄山市五城镇林区的马尾松和杉木进行识别。通过对Hyperion影像进行一阶、二阶微分变换,优化组合487～559 nm和681～742 nm光谱范围中反射差异明显的波段,再结合支持向量机(support vector machine,SVM)模型进行林种间分类识别。基于Hyperion影像像元反射率及其一阶和二阶微分光谱的分类识别总体精度分别达到76. 50%,81. 42%和88. 52%,对应Kappa系数分别为0. 528 4,0. 625 7和0. 769 1。结果表明,基于二阶微分变换的高光谱数据,通过SVM模型,可有效提高马尾松和杉木的识别精度,为高光谱遥感针叶林种分类识别提供了一种技术途径。

Hyperspectral remote sensing can distinguish small spectrum differences between ground objects,and is expected to solve the classification problem of tree species. In this paper,by using Hyperion hyperspectral image,combined with the ground measured samples,classification of Pinus massoniana and Cunninghamia lanceolata in Wucheng of Huangshan City was conducted. With the 1 st and 2 nd differential transformation of the image,spectral band combination of 487 ~ 559 nm and 681 ~ 742 nm differs significantly,and hence was chosen to conduct supervised classification using support vector machine. Classification accuracy of raw,1 st and 2 nd differential transformation image reaches 76. 50%,81. 42% and 88. 52% with Kappa coefficient being 0. 528 4,0. 625 7 and0. 769 1 respectively. The results show that 2 nd differential transformation and band selection of hyperspectral data can improve the classification accuracy of Pinus massoniana and Cunninghamia lanceolata,thus providing a foundation for further study of classification of coniferous forest with hyperspectral remote sensing.