摘要
湿地是生态系统的重要组成部分,及时、准确地获得湿地基础信息,对湿地的动态监测、保护与可持续利用及其它领域的研究具有重要意义。以三江平原东北部沼泽湿地为例,利用分类回归树(Classification and Regression Tree,CART)算法从训练样本数据集中挖掘分类规则,集成遥感影像的光谱特征、纹理特征和地学辅助数据建立研究区湿地信息提取的决策树模型。用实测的GPS样本点对分类结果进行精度验证,并与最大似然监督分类方法(Maximum Likelihood Classi-fication,MLC)进行对比。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82.65%和0.7935,分类精度较MLC监督分类方法有明显提高,是内陆淡水沼泽湿地信息提取的有效手段。
Wetlands are considered an integral part of the global ecosystem. Enhancement of their scientific management informed by quantitative, accurate and repeatable observations of wetlands＇ landscape would obviously be significant. Taking the northeast of the Sanjiang Plain as a case study, we use classification and regression tree （CART） algorithm for purposes of mining classification rules from training samples. Classification tree model of wetland information extraction was built from these samples through CART algorithm,which integrates spectral ,texture and the assistant geographical characteristics. The classification results based on CART algorithm were checked by statistical confusion matrix accuracy assessment using field GPS samples. Validation shows that total classification accuracy is 82.65%, Kappa coefficient is 0. 7935. The results had suggested that the accuracy of classification based on the CART algorithm was higher than the MLC supervised classification method. The developed method is portable,relatively easy to implement ,and should be applicable in other settings and larger extents.
出处
《遥感技术与应用》
CSCD
2008年第4期365-372,I0001,共9页
Remote Sensing Technology and Application
基金
国家“十一五”科技支撑重点项目(2006BAD23B03)
中国科学院知识创新工程重要方向项目(KZCX3-SW-356)
联合国开发计划署(UNDP)/全球环境基金(GEF)项目(CPR/98/G32)
地球系统科学数据共享网项目(2006DKA32300-16)
关键词
湿地
遥感
信息提取
决策树模型
三江平原
Wetland
Remote sensing
Extraction of information
Decision tree model
Sanjiang plain
