MODIS土地利用/覆被多时相多光谱决策树分类  

Land use/cover decision tree classification fusing multi-temporal and multi-spectral of MODIS

作　　者：刘建光[1,2] 李红[2] 孙丹峰[1] 张微微[2] 周连第[2] 

机构地区：[1]中国农业大学资源与环境学院,北京100193 [2]北京市农林科学院农业综合发展研究所,北京100097

出　　处：《农业工程学报》2010年第10期312-318,389,共8页Transactions of the Chinese Society of Agricultural Engineering

基　　金："十一五"国家科技支撑项目(2006BAB15B05);北京市财政局项目支持

摘　　要：利用MODIS多时相与多光谱结合,尝试探讨低成本、高精度的北京土地利用/覆被实时获取方法。首先根据归一化植被指数(NDVI)的均值、标准差建立了研究区各地类的典型NDVI时间序列曲线,进而提取了6个可以反映区域物候模式、植被生长速率等信息的分类参数;然后对反映地表土壤信息较多的3月份多光谱影像进行主成分变换,选取第一主成分(PC1)作为辅助分类参数;最后基于分类回归树(CART)算法进行监督决策树分类。经SPOT-5影像验证,分类总体精度达到83%,Kappa系数为0.769,PC1辅助分类后总体精度提高了2.1%。研究表明:建立的典型NDVI序列曲线具有一定的区域地类代表性,并且提取的分类参数具有较强的地类分异性;3月份MODIS多光谱影像可以有效提高MODIS-NDVI时间序列分类的精度。To explore the low-cost,high-precision real-time access method of land use/cover using the combination of MODIS multi-temporal and multi-spectral is very necessary for quickly assess regional land use/cover change.Firstly,using land use/cover types samples＇ mean and standard deviation,the study established each types＇ typical NDVI time series curves in the Beijing study area,then extracted 6 classification parameters quantifying phenophase types,agricultural cultural patterns and plants growth rate etc.Secondly,principal component transform was performed to March month multi-spectral remote sensing image,and the PC1（first principal component） was selected as an assistant classification parameter that can reflect more bare soil alike information.At last,land use/cover was classified in CART（classification and regression tree） decision tree method integrating the above types parameters.Classification accuracy was tested by SPOT-5 image,overall accuracy is 83%,Kappa coefficient is 0.769,the classification overall accuracy with PC1 is 2.1% higher than without it.The results show that：the established typical NDVI time series curves have a strong presentation of land use/cover types in this region,classification parameters extracted are very distinguishing in types.March month MODIS multi-spectral imaging can efficiently improve the MODIS-NDVI time series＇ classification accuracy.

关 键 词：土地利用 主成分分析 分类 MODIS 精度检验 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]