基于CART决策树分类的江苏省湿地提取
Extraction of Remotely Sensed Wetland Information for Jiangsu Province Based on CART Decision Tree Classification
投稿时间:2020-03-09  修订日期:2022-05-20
中文关键词:湿地类型  CART决策树  特征指数  纹理信息
英文关键词:wetland classification  CART decision tree  characteristic index  textural features  
基金项目:江苏省测绘地理信息科研项目(JSCHKY201806)
作者单位
冯婉玲 南京林业大学土木工程学院江苏 南京 210037 
何立恒 南京林业大学土木工程学院江苏 南京 210037 
杨 强 南京林业大学土木工程学院江苏 南京 210037 
摘要点击次数: 174
全文下载次数: 54
中文摘要:
      针对湿地环境复杂、类型多样且难以从景观尺度进行识别的特点,为解决遥感影像“同物异谱、异物同谱”的难点,探讨湿地遥感信息提取方法,可为湿地管理保护提供基础数据。以江苏省为研究区,高分一号影像为数据源,结合不同地物的环境特征和空间特征,采用CART决策树法进行地物分类,并提取湿地信息;选取影像纯净像元为训练样本,根据样本数据特征,制定分类规则。集成高分影像的光谱特征、植被指数、水体指数、土壤调整植被指数、纹理信息、主成分分析波段和辅助地形数据构建CART决策树模型,实现地物分类,最终采用混淆矩阵和分类精度指标进行评价,并与最大似然分类方法进行对比。结果表明,分类总体精度达到86.77%,Kappa系数为0.85,精度较最大似然分类提高了将近16%,Kappa系数提高了近0.2;遥感解译成果统计表明,2016年江苏省湿地总面积为14053.12 km2;其中,水域面积为12585.59 km2,沼泽574.18 km2,滩涂893.35 km2,分别占湿地总面积的89.56%、4.09%、6.35%。研究发现,CART决策树分类精度高于最大似然监督分类,具有较强实用性和优越性,对大面积湿地信息提取具有借鉴意义。
英文摘要:
      Wetland environments are complex and diverse, making it difficult to identify features at the landscape scale using remote sensing image technology. To solve the difficulty of “same objects with different spectra and different objects with the same spectrum”, we explored extraction of wetland remote sensing information using the CART (Classification And Regression Tree) decision tree method. The results provide basic data for wetland management and protection, as well as technological support for remote sensing of wetland conditions. Jiangsu Province was the study area and GF-1 images from 2016 was the data source. Environmental and spatial characteristics of different physical features were used to identify landscape types and wetland information was extracted using CART. First, pure pixels were selected as training samples, and classification rules were formulated according to the characteristics of sample data. The CART decision tree model was constructed by integrating the spectral characteristics of high-resolution images, vegetation index, water index, soil adjusted vegetation index, texture information, waveband by principal component analysis and auxiliary terrain data. Then the CART model was used to classify landscape type. Finally, the confusion matrix and classification accuracy index were used to evaluate the accuracy of classification results, and the results were compared with those obtained by the MLC (maximum likelihood classification) method. The overall accuracy of the classification reached 86.77%, with a Kappa (κ) coefficient of 0.85. Compared with the maximum likelihood classification, CART increased accuracy by nearly 16%, and the Kappa coefficient by nearly 0.2.The statistics obtained by interpreting remotely sensed images show that the total area of wetlands in Jiangsu Province in 2016 was 14 053.12 km2 and included 12 585.59 km2 of water (89.56%), 574.18 km2 of swampland (4.09%) and 893.35 km2 of tidal flats (6.35%). In conclusion, the accuracy of CART decision tree classification is higher than for MLC, indicating that the CART decision tree method is better for extracting information on large area wetlands from satellite imagery.
冯婉玲,何立恒,杨 强.2022.基于CART决策树分类的江苏省湿地提取[J].水生态学杂志,43(3):35-43.
查看全文  查看/发表评论  下载PDF阅读器
Copyright © 2008 《水生态学杂志》编辑部 地址:湖北省武汉市雄楚大街578号 
邮编:430079 电话:027-82926630 E-mail: sstx@mail.ihe.ac.cn  京ICP备09084417号