基于GEE云平台和Sentinel数据的滇池湿地制图  

Wetlands Mapping in Dianchi Lake Based on GEE Cloud Platform and Sentinel Data

作　　者：罗蓉蓉 董燕[1] LUO Rong-rong;DONG Yan(Kunming University of Science and Technology,Kunming,Yunnan 650093)

机构地区：[1]昆明理工大学国土资源工程学院,云南昆明650093

出　　处：《安徽农业科学》2025年第4期205-208,242,共5页Journal of Anhui Agricultural Sciences

摘　　要：基于Google Earth Engine(GEE)和Sentinel数据,结合地形数据,提取影像的光谱指数、红边指数、纹理特征、雷达特征和地形特征,通过RF-RFE方法筛选特征得到最优特征数据集,使用基于像元的方法(随机森林)和面向对象的方法(简单非迭代聚类+随机森林)实现滇池湿地制图,探讨不同分类方法、特征变量对滇池湿地制图的影响。结果表明,面向对象方法优于基于像元方法,总体精度为90.86%,Kappa系数为0.887。面向对象方法可以有效减轻“椒盐现象”,以及湿地和非湿地错分现象,RF-RFE方法可以去除冗余特征,有效提高分类效率。基于GEE平台的面向对象方法适合于高原湖泊滇池湿地制图。Based on Google Earth Engine(GEE)and Sentinel data,combined with terrain data,the spectral index,red edge index,texture features,radar features and terrain features of the images were extracted.The RF-RFE method was used to select features to obtain the optimal feature dataset,and Pixel based methods(random forest)and object-oriented methods(simple non iterative clustering+random forest)were used to map the Dianchi Lake wetland.We explored the impact of different classification methods and feature variables on the mapping of Dianchi Lake wetlands.The results indicated that the object-oriented classification method outperformed the pixel-based classification method,with an overall accuracy of 90.86%and a Kappa coefficient of 0.887.The object-oriented method was effective in minimizing the‘salt and pepper'phenomenon and the misclassification of wetlands and non-wetlands,while the RF-RFE method could remove redundant features and effectively improve classification efficiency.The object-oriented method based on the GEE platform was suitable for mapping the wetlands of highland Dianchi Lake.

关 键 词：滇池 湿地 随机森林 简单非迭代聚类 Google Earth Engine SENTINEL 

分 类 号：S127[农业科学—农业基础科学]