基于ETM影像多种特征的CART决策树分类——以延边州为例

CART-based on Decision Tree Classifier Using Multi-feature of ETM Image——A Case Study in Yanbian Area

分类回归树(CART)是一种非参数化的分类与回归方法,在用于遥感影像自动分类时,可方便地应用多源知识,提高分类精度。以延边州试验区土地利用/覆被分类为例,利用分类回归树分析从训练样本中集中发现分类规则,集成遥感影像的光谱特征、纹理特征和辅助地学特征进行分类试验,并与传统的最大似然分类方法进行比较。结果表明,基于CART的决策树分类结果的总精度和Kappa系数分别为90.37%和0.8863,分类精度比MLC监督分类方法有明显提高。

CART was a kind of non- parameter classification and regression method, which used multi - source data to improve classification accuracy when used in automatic classification based on remote sending images. Taking Yanbian area in Jilin Province as a ease study, classification and regression tree （CART） was used to classify the ETM image. Classification tree model integrated spectral, texture and the assistant geographical charaeteristics. The results showed that the accuracy of classification based on the CART was higher than the MLC supervised classification method. And its total classification accuracy was 90.37%, Kappa coefficient was 0. 8863.