基于MODIS影像多特征的CART决策树分类

以山东省为研究区域,利用2009年9月MODIS的8 d合成波段反射率产品MOD09,选择特征变量植被指数(NDVI、EVI)、NDWI、NDMI、NDSI及辅助信息DEM,通过选取其中的影像特征组合来确定分类方案,构建各波段组合的CART决策树,对MODIS影像进行分类,得到CART决策树的最优波段组合。结果表明,特征变量DEM、NDVI、EVI对分类结果贡献较大;将CART决策树的分类结果与其相对应的最大似然分类结果进行比较可知,基于影像多特征的CART决策树分类方法能明显提高分类精度。

CART Decision Tree Classifier Based on Multi-feature of MODIS Data

Taking Shandong Province as the study area,we chose composite albedo MODIS products MODIS09Q1(B1～B2 band in September 2009,250 m resolution)、MODIS09A1(B3～B7 band,500 m resolution) for one period of 8-day,characteristics variables vegetation index(NDVI,EVI,)NDWI,NDSI,and auxiliary information DEM by selecting a combination of image features to determine the classification schemes.The CART decision tree was built for each kind of band combination to classify MODIS images.The optimum band combination of the CART decision tree was composed of the bands of B1～B7,DEM,NDVI,NDMI and Feature variables DEM,NDVI,EVI make a greater contribution to classification results.Comparing CART decision tree classification results with their corresponding maximum likelihood classification results,it show that the CART decision tree classification based on image features can significantly improve the classification accuracy.