基于资源一号02C卫星数据的面向对象的森林植被分类

摘要

ZY1-02C high resolution satellite is research and developed by our country .In this paper, we do research on forest classi-fication use object-based information extraction technology with ZY 1-02C high resolution image .Firstly, implement multi -scale segmentation and obtain the optimal scale;secondly , developing hierarchical structure considering the object features and relationship of each other;then, extract land cover informationwith the help of a variety of object characteristics like spectrum , texture, shape and so on.Our study area is Guangxi Mao′er Mountain Nature Reserve , classify the study area into seven land cover types include conifer-ous forest, broad-leaved forest, bamboo, shrub, farmland, non-vegetation area and shade , the overall accuracy is 82.51%, kappa coefficient is 0.79, higher than the object -based Nearest Neighbor classification and pixel -based Maximum Likelihood classifica-tion.%资源一号02 C卫星是我国自主研发的高分辨率卫星。利用面向对象的信息提取技术，开展基于资源一号02C高分辨率数据的林区植被分类，具体分为三个步骤：1）对影像进行多尺度分割，获取最优尺度；2）根据各类地物特点及相互间关系，建立地物类型层次；3）结合光谱、纹理、形状多种对象特征，进行地物分类。以广西猫儿山自然保护区为例，根据区内地物特点，将地物分为针叶林、阔叶林、竹林、灌丛、耕地、非植被、阴影等7种类型，经检验表明该方法总体分类精度达到82．24％，kappa系数为0．77，优于面向对象的最邻近法和基于像元的最大似然分类。