结合灰度纹理和支持向量机分类的三江源草地信息提取  

Information Extraction of Sanjiang Source Grassland Based on Gray Level Texture and Support Vector Machine Classification

作　　者：王爱芳[1] 王妮[1,2] 

机构地区：[1]滁州学院地理信息与旅游学院,安徽滁州239000 [2]安徽省地理信息集成应用协同创新中心,安徽滁州239000

出　　处：《安徽农业科学》2017年第13期210-213,231,共5页Journal of Anhui Agricultural Sciences

基　　金：安徽省高校省级自然科学研究重点项目(KJ2016A531)

摘　　要：以三江源地区地形地貌特征、草场分布较为典型的班玛县为例,以HJ环境星多光谱影像为主要数据,基于支持向量机SVM超平面理论,结合灰度共生矩阵寻找最适宜的分类核函数,选取了三江源地区草地信息提取的最适宜SVM分类模型,并与传统的监督分类方法最大似然法和SVM分类方法进行比较,进行三江源草地分类方法的优化。结果表明,与传统监督分类方法相比,除Sigmoid核函数外,其余结合方法的分类精度均有所提高,其中结合纹理和高斯核函数的SVM分类模型有着较理想的识别效果,精度达到91%,Kappa系数为0.856 0,能为三江源地区草地可持续利用以及生态系统恢复提供基础数据。Taking Banma County with typical landform characteristics and distribution of grassland in Sanjiang Source as an example, HJ environmental satellite multispectral images were used as main data. Based on the SVM super plane theory, combined with gray level co-occurrence matrix, the most suitable classification kernel function was discussed. The optimum SVM classification model of grassland information extraction in the source region of Sanjiang was selected and compared with the traditional supervised classification methods（ maximum likelihood method and SVM classification method）. The classification method of grassland in Sanjiang source was optimized. The results showed that the classification accuracy of other methods was improved compared with the traditional supervised classification method, in addition to Sigmoid kernel function. SVM classification model combined with texture and Gauss kernel function had ideal recognition effects, the accuracy was 91% , Kappa coefficient was 0. 856 0. The research results can provide basic data for the sustainable utilization of grassland and the restoration of ecosystem in the source region of Sanjiang.

关 键 词：三江源 草地信息提取 遥感 支持向量机 灰度纹理 

分 类 号：S127[农业科学—农业基础科学]