多维数据复合下QUEST决策树遥感影像分类——以九江市为例

Remote Sensing Image Classification of QUEST Decision Tree Based on Multi-dimensional Data——A Case Study of Jiujiang City

遥感影像分类是信息提取的关键技术,融合多种类型的数据可以提高影像的解译能力。以九江市Landsat 5TM影像为实验对象,采用多维数据复合下快速、无偏、高效统计树（quick unbiased efficient statistical tree,QUEST）算法的决策树分类方法,综合影像的光谱特征、地理信息数据和影像分类结果,根据样本数据自动挖掘分类规则,实现影像解译,并将分类结果和普通的QUEST决策树分类进行定量比较分析。结果表明：（1）采用多维数据复合下QUEST决策树分类的总精度和Kappa系数分别为89.83%和0.872,较普通的QUEST决策树分类方法有较大提高;（2）在地物类型丰富的区域,多维数据复合下QUEST决策树分类方法适应性更好;（3）选用SVM（support vector machine）分类结果作为辅助数据,滩地和水库坑塘的用户精度提高最为明显,分别提高了29.96%和17.08%。

Taking Jiujiang city as the study area,this paper proposes a method of quick unbiased efficient statistical tree（QUEST）decision tree classification based on multi-dimensional data.Combining the image spectral characteristics,geographic data and image results,according to automatic data mining classification rules,the image interpretation is realized and the quantitative analysis is carried out by comparing classification results with the ordinary QUEST decision tree.We find out that：（1）The overall accuracy and Kappa coefficient of QUEST decision tree classification based on multi-dimensional data are 89.83% and 0.872,the classification method has been greatly improved compared with ordinary QUEST decision tree.（2）This classification method is more adaptable in the area of feature type.（3）Taking support vector machine classification results as the assistant data,the user accuracy of the reservoir of pits and beach wetlands increase by 29.96% and 17.08% respectively.