基于CART决策树的黄河源区域生态地类的遥感解译研究

Research on Remote Sensing Interpretation of Ecological Land Types in the Source Region of the Yellow River Based on CART Decision Tree

为指导黄河源区域生态环境修复,以黄河源区域为研究区,采用Landsat 8影像作为数据源,以融合多种特征指数的CART树模型进行高寒湿地分类方法研究。结果表明:使用CART决策树算法进行数据挖掘处理,将得到的决策树用来分类制图,得到黄河源地区生态地类分布图,通过混淆矩阵得出分类后总精度为88.25%,Kappa系数为0.8345,而使用同样分类样本的监督分类方法中使用范围较广的最大似然法分类得到的总体精度为84.90%,Kappa系数为0.7888,分别低于CART决策树分类3.35百分点与0.0457,并且各地类的精度均低于CART树分类方法,证明本研究所构建的决策树分类模型适用于研究区的生态地类提取。

To guide the restoration of the ecological environment in the Yellow River source region,the alpine wetland classification study was conducted using Landsat 8 imagery as the data source and a CART number model incorporating multiple feature indices,using the Yellow River source region as the study area.The results showed that the data mining process using CART decision tree algorithm and the obtained decision tree were used for classification and mapping to obtain the distribution map of ecological land types in the Yellow River source area,and the total accuracy after classification was 88.25%with a Kappa coefficient of 0.8345 by confusion matrix,while the overall accuracy obtained by the supervised classification method using the same classification sample with a wider range of maximum likelihood was 84.90%and the Kappa coefficient was 0.7888,3.35 percentage points and 0.0457 were lower than the CART decision tree classification,and the accuracy of each class was lower than the CART tree classification method.