基于Landsat OLI影像的盐城滨海湿地分类方法对比研究  

Comparisons of Classification Methods for Coastal Wetlands in Yancheng City Based on Landsat OLI Images

作　　者：刘翔[1,2,3,4] 刘会玉 林振山[1,2,3,4] 贾俊鹤 吕士成[5] 

机构地区：[1]南京师范大学地理科学学院,江苏南京210023 [2]虚拟地理环境教育部重点实验室,江苏南京210023 [3]江苏省地理环境演化国家重点实验室培育建设点,江苏南京210023 [4]江苏省地理信息资源开发与利用协同创新中心,江苏南京210023 [5]江苏盐城自然保护区管理处,江苏射阳224333

出　　处：《湿地科学》2017年第5期689-696,共8页Wetland Science

基　　金：江苏省自然科学基金项目(31470519和31370484);江苏省高校优势学科建设工程项目(BK20131399)资助

摘　　要：采用何种方法对滨海湿地信息进行有效提取和分类是值得关注的问题。利用Landsat OLI影像数据,在ENVI 5.1软件平台上,采用监督分类、决策树分类和面向对象方法,对盐城滨海湿地进行了分类研究。研究结果表明,在分类精度和效果上,面向对象方法的分类结果最好,其它依次为决策树分类方法、监督分类方法的分类结果;3种方法对光滩和芦苇(Phragmites australias)滩涂信息的提取效果相对较好,对道路和碱蓬(Suaeda glauca)滩涂信息的提取效果相对较差;面向对象方法能够克服传统基于像元分类方法中的"椒盐效应"问题,并能取得较高的分类精度,该方法适用于利用中分辨率遥感影像的滨海湿地分类研究。Finding an appropriate method for the effective extraction and accurate classification of coastal wetlands is a necessary step for the implementation of effective ecological management strategies. Therefore,based on the Environment for Visualizing Images（ENVI） software platform and Landsat OLI data, the performances of three methods（supervised classification, decision tree classification and object-oriented classification methods） for classification of the coastal wetlands were investigated. The results showed that object-oriented classification outperformed decision tree classification and supervised classification in terms of discriminative power. Meanwhile, all the methods exhibited better discriminability for mudflat and Phragmites australis tidal flat and yielded higher classification accuracy, however, the discriminability for roads and Suaeda salsa tidal flat generated from those methods were relatively poorer. The object-oriented method not only could well overcome the ＂salt and pepper＂ effect of traditional pixel-based classification methods, but could improve the classification accuracy greatly. Thus, this method could provide important technical support and reference for classification of the coastal wetlands based on the medium resolution remote sensing images.

关 键 词：滨海湿地 LANDSAT 监督分类 决策树分类 面向对象分类 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]