基于高分辨率遥感影像多维度特征的树种识别方法研究

Study on Tree Species Recognition Methods Based on Multi-Dimensional Feature of High Resolution Remote Sensing Image

以北京市西山试验林场为研究区域,利用Worldview—2影像构建各树种的光谱特征、地形特征、植被指数特征、纹理特征以及形态特征,建立关于山地森林树种识别的知识。采用基于像元和面向对象的方法进行树种识别分类。在基于像元的分类方法中,选择决策树分类和支持向量机分类;在面向对象的分类方法中,选择基于边缘检测的方法分割影像,用最近邻法分类。决策树分类的总体分类精度为65.62%,Kappa系数为0.588 9;支持向量机分类的总体分类精度为62.42%,Kappa系数为0.552 8;面向对象的分类方法总体分类精度为64.27%,Kappa系数为0.580 2。

Taking the Xishan Experimental Forest Farm in Beijing as the research area , we established the spectral features, topographic features, vegetation index features, texture features and morphological features of the tree species based on the Worldview-2 remote sensing data, established knowledge about mountain forest tree species recognition. Image classification based on pixel and object-oriented methods were used in this stud- y. In the pixel-based classification method, decision tree classification and support vector machine classification were chosen. In the object-oriented method, the image was segmented by edge detection and classified by k- nearest neighbor method. The overall classification accuracy of the decision tree was 65.62% and the Kappa coefficient was 0. 588 9. The overall classification accuracy of the support vector machine was 62. 42% and the Kappa coefficient was 0. 552 8. The overall classification accuracy of the object-oriented method was 64.27% and the Kappa coefficient was 0. 580 2.