基于卡方自动交互检测的GF-1影像鄱阳湖湿地信息提取  

Extraction of Wetland Information in Lake Poyang from GF-1 Images Based on CHi-squared Automatic Interaction Detection

作　　者：李婷 薛丽 顾华奇 LI Ting;XUE Li;GU Huaqi(Jiangxi Geomatics Center,Nanchang 330209,China;Key Laboratory of Poyang Lake Wetland and Watershed Research,Ministry of Education,Jiangxi Normal University,Nanchang 330022,China)

机构地区：[1]江西省基础地理信息中心,江西南昌330209 [2]江西师范大学鄱阳湖湿地与流域研究教育部重点实验室,江西南昌330022

出　　处：《测绘与空间地理信息》2018年第9期68-70,74,共4页Geomatics & Spatial Information Technology

基　　金：江西省重大生态安全问题监控协同创新中心专项(JXS-EW-012);国家科技支撑计划课题(2015BAH50F01)资助

摘　　要：快速准确地获取鄱阳湖区域湿地信息,对湿地进行动态监测具有重要价值。本文以国产高分一号(GF-1)影像和辅助地学数据作为数据源,将GF-1影像获取的归一化植被指数(Normalized Differential Vegetation Index,NDVI)、基于蓝光波段的归一化差异水体指数(Normalized Difference Water Index based on Bule light,NDWI-B)以及数字高程模型(Digital Elevation Model,DEM)作为分类变量,采用卡方自动交互检测(CHi-squared Automatic Interaction Detection,CHAID)决策树算法构建分类规则,并进行研究区范围内的湿地信息提取。最后,采用混淆矩阵对分类结果进行精度验证,并与最大似然法监督分类结果进行比较。结果表明,分类结果的总体精度和Kappa系数分别为85.6%和0.82,较最大似然法监督分类结果分别提高了9.6%和0.12,是鄱阳湖区域湿地信息提取的有效方法。It is of great value to quickly and accurately obtain wetland information and dynamically monitor wetland of Poyang Lake area. This study chose GF-1 image and auxiliary geosciences data as the data source, with the normalized differential vegetation index （NDVI） , normalized difference water index based on bule light （NDWI-B） and digital elevation model （DEM） as categorical variables, then applied the decision tree algorithm of CHi-squared automatic interaction detection （CHAID） to construct the classification rules, which used to extract wetland information. Finally, we used confusion matrix to verify the accuracy of classification results, compared with the result of supervise classification based on the method of maximum likelihood. The results obtained overall accuracy of 85.6% and Kappa coefficient of 0.82, compared with the method of maximum likelihood, the supervised classification results were improved by 9.6% and 0.12, which is an effective method for extraction of wetland information in Poyang Lake region.

关 键 词：鄱阳湖 GF-1 决策树 卡方自动交互检测 湿地信息提取 数字高程模型 

分 类 号：P231[天文地球—摄影测量与遥感]