GF-2影像多特征组合的毛竹林信息精细提取

Phyllostachys pubescens Forest Detection for Gaofen-2 Satellite Image Data Using Multiple Features

以福建省永安市天宝岩山地毛竹林为对象,基于高分二号(GF-2)卫星数据提取光谱、纹理及物候特征,比较光谱+纹理、光谱+纹理+物候2种特征组合方案,采取支持向量机(SVM)、最大似然法(MLE)、随机森林(RF)等3种分类方法对毛竹林信息进行专题提取。研究表明:4月份不同森林类型间的光谱特征差异大,是毛竹林信息提取的关键物候期,基于GF-2的10月与4月归一化植被指数(NDVI)的差值、乘积运算构建一种新的叶物候特征指数以适用毛竹林信息的分类提取;增加叶物候特征后有利于提高毛竹林专题信息分类精度,SVM、MLE、RF三种方法的总体分类精度分别提高了0.69%、0.54%、0.63%,Kappa系数达0.79~0.84,其中支持向量机的分类结果最好,对毛竹林信息提取精度达92.75%。

Taking moso bamboo forest in Tianbaoyan mountainou area from Yong’an City, Fujian Province as the detection object, spectrum, texture and leaf phenological features were extracted by Gaofen-2 remote sensing data(GF-2).Then, two features combination schemes, spectrum and textures, spectrum and texture and phenology were compared. The Support vector machine(SVM), maximum likelihood method(MLE) and Random forest(RF) methods were adopted to extract the spatial distribution of thematic bamboo forest. The research results showed that there are great differences in spectral characteristics among four forest types in April, which is the key phenological period for information extraction of Moso bamboo forest. A new leaf phenological characteristic Index for bamboo forest was proposed based on the difference and product by the two NDVI values from the October and April GF-2 data. The leaf phenology characteristics was beneficial to improve the classification accuracy for the bamboo forests. The overall classification accuracy of SVM, MLE, and RF methods were improved by 0.69%, 0.54% and 0.63%, and the kappa coefficient were 0.79,0.79 and 0.84, respectively. The optimal classification method was SVM by mean of assessing the classification accuracy and kappa coefficient, and the mapping accuracy of Moso bamboo forest was 92.75%.