基于水稻特征波段的决策树分类研究

以滁州市为例,结合水稻物候的特征波段,选用反映水稻物候期时相的TM数据,并基于多特征波段,构建CART决策树分类提取水稻种植面积。结果表明,植被指数、湿度因子、绿度因子、纹理特征等多特征参与CART决策树分类能够提高总体精度。基于光谱信息、植被指数和纹理特征的决策树分类的总精度比以最大似然法进行的监督分类方法提高了6.942 1百分点,Kappa系数提高了0.110 4。合理选用作物物候期数据及其遥感影像的特征波段能够有效降低分类误差,为地形复杂地区获取作物种植面积提新的方法。

Study on Decision Tree Classification Based on Multi-temporal Characteristic Bands of Rice

Taking Chuzhou City as an example, combined with the characteristic band of rice phenology in remote sensing image, using TM data of the typical rice phenology, planting area of rice was extracted by CART decision tree classification based on the multi-feature band.The results showed that CART algorithm classification involved by vegetation index, humidity factor,green degree factor,texture feature and other characteristics can improve the overall accuracy.The decision tree classification based on spectral information, vegetation indices and texture features was 6.9421 percentage point higher in overall accuracy than that of supervised classification method based on maximum likelihood, and its Kappa coefficient increased 0.1104.So the reasonable selection of the characteristic bands of crop phenological data and remote sens-ing images could effectively reduce the classification error, which could provide new method for obtaining crop planting area extraction in the complex terrain area.