基于决策树C4.5算法的面向对象分类方法研究  

Remote Sensing Image Classification Based on Object-oriented Technique and C4.5 algorithm

作　　者：陈丽萍[1] 武文波[2] 

机构地区：[1]铜陵学院土木建筑系,铜陵244000 [2]辽宁工程技术大学测绘与地理科学学院,阜新123000

出　　处：《遥感信息》2013年第2期116-120,共5页Remote Sensing Information

基　　金：辽宁省教育厅创新团队项目计划(LT2010043)

摘　　要：提出了基于决策树C4.5算法的面向对象分类方法。首先利用多尺度分割方法对遥感影像进行分割;然后提取对象的特征信息,包括光谱特征、纹理特征、形状特征、图层特征等,对提取的众多的对象特征,利用决策树C4.5算法对其进行知识挖掘,自动建立分类规则;最后利用建立的分类规则,将C4.5算法作为一种分类器对分割后的遥感影像进行分类,并以南充市城市土地利用为例进行了分类实验。实验验证了该方法的可行性。实验结果表明利用决策树C4.5算法建立的分类规则准确率高,利用该分类规则进行的面向对象分类效果较好。An object-oriented remote sensing image classification based on C4.5 algorithm was proposed in this paper. First-ly, the image was segmented into image objects using the multi-scale segmentation method. Then the features of objects, such as spectral, texture,shape and layer characteristics,were measured. The rules were automatically obtained using C4.5 algorithm. At last,image classification was carried out using the rules. The accuracy of the proposed approach was checked by a case study in Nanchong,China. Experimental result showed that the rules obtained using the C4.5 algorithm have high accuracy,and the result of the classification using the obtained rules is preferable.

关 键 词：遥感 面向对象 C4 5算法 规则 分类 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]