基于GMM与图像多元特征的自动决策树分类方法研究  

Automatic Decision Tree Classification Method Based on GMM and Multi-features

作　　者：赵健赟[1,2] 彭军还[1] ZHAO Jian-yun PENG Jun-hua(School of Land Science and Technology, China University of Geosciences, Beijing 100083, China Department of Geologic Engineering ,Qinghai University ,Xining 810016 ,China)

机构地区：[1]中国地质大学(北京)土地科学技术学院,北京100083 [2]青海大学地质工程系,青海西宁810016

出　　处：《内蒙古师范大学学报（自然科学汉文版）》2016年第5期709-715,722,共8页Journal of Inner Mongolia Normal University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(41374016);青海高原北缘新生代资源环境重点实验室建设项目(2012-Z-Y14)

摘　　要：通过对Landsat 8OLI影像空间和光谱特征的分析,使用GMM描述各地物的分布特征,并利用EM算法估计其参数.在获得影像GMM贝叶斯分类信息的基础上,融合主成分、相异性纹理、FNDWI、NDBI和NDVI等其他多元特征,自动构建CART决策树对图像进行分类.结果表明,该方法的总体分类精度比其他方法最大提高3.82%,比利用影像其他特征的分类精度最大提高9.78%,而高斯混合模型信息的融合可显著提高林地、耕地等地物分类的生产精度和用户精度.GMM was employed to descript the land types,and EM algorithm was used to estimate its parameter through the analysis of the image＇s space and spectral features.On the basis of obtained GMM Bayesian classification information of the image and fused its principal component,diversity texture,FNDWI,NDBI,NDVI and other multi-features,A CART decision tree was automatically constructed to classification the image of experimental area.The research shows that the method proposed in this paper could increase 3.82% overall classification accuracy compared with other methods,and increase 9.78%classification accuracy compared with other features combination,and the use of GMM improved the production and user classification accuracy of the forest and cultivated land obviously.

关 键 词：自动决策树 GMM FNDWI EM算法 多元特征 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]