一种改进的SVM决策树及在遥感分类中的应用  

Improved SVM decision-tree and its application in remote sensing classification

作　　者：丁胜锋[1,2] 孙劲光[1] 陈东莉[3] 李扬[1] 姜晓林[2] 

机构地区：[1]辽宁工程技术大学,辽宁葫芦岛125105 [2]辽宁石油化工大学经济管理学院,辽宁抚顺113001 [3]中国石油抚顺石化公司石油二厂科技信息部,辽宁抚顺113004

出　　处：《计算机应用研究》2012年第3期1146-1148,1151,共4页Application Research of Computers

基　　金：辽宁省科技计划资助项目(2010401010)

摘　　要：针对遥感图像分类问题提出了一种基于遗传算法和K近邻的SVM决策树方法。算法以基于类分布的类间分离性测度为准则,利用遗传算法对传统的SVM决策树进行优化,生成最优(较优)决策树。在分类阶段,对容易分的节点利用SVM进行分类,而对可分离性差的节点采用SVM和K近邻相结合的分类方法,最终实现多类别分类。实验结果表明,与传统的分类方法相比,该算法的实验效果较好,可有效地提高遥感图像的分类精度。This paper presented a SVM decision-tree algorithm based on GA and KNN. First, GA was used to create optimal or near-optimal decision-tree, which defined a novel separability measure. Then in the class phase, standard SVM was used to make binary classification for the divisible nodes, and SVM combined with KNN were used to classify the fallible nodes. Finally, achieved the multi-classification by the SVM decision-tree. Experimental results show that the proposed method can effectively improve the classification precision of remote sensing image in comparison to traditional classification methods.

关 键 词：遗传算法 K近邻 支持向量机决策树 遥感图像分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]