摘要
基于Sentinel-2数据,以盐城国家级珍禽自然保护区核心区为研究区,采用基于面向对象的随机森林模型对研究区内的湿地信息进行分类研究.首先,对影像进行分割处理,计算光谱特征、纹理特征、水体指数、植被指数与纹理特征,并对特征重要性进行排序筛选.其次,基于此构建5种特征组合方案,并对研究区进行分类,比较不同组合的分类精度找出研究区最优的特征组合方案.最后,实验表明:通过特征优选后的随机森林算法进行分类效果最好,总体精度达到87.07%,Kappa系数为0.84.其中互花米草在3种植被中分类精度最高,为97.73%.证明此方法能够有效提高滨海湿地的分类精度,可用作该区域湿地变化研究.
Based on Sentinel-2 data,this paper uses the core area of Yancheng National Rare Bird Nature Reserve as the study area,and uses an object-oriented random forest model to classify the wetland information in the study area.Firstly,the images are segmented,based on which the spectral features,texture features,water body index,vegetation index and texture features are calculated and ranked;based on this,five feature combination schemes are constructed and the study area is classified,and the best feature combination scheme is found by comparing the classification accuracy of different combinations.The results show that the random forest algorithm after feature selection is the best,with an overall accuracy of 87.07%and a Kappa coefficient of 0.84.Among the three vegetation species,the classification accuracy of Spartina Alterniflora is the highest at 97.73%,which proves that this method can effectively improve the classification accuracy of coastal wetlands and thus can be used for the study of wetland change in the region.
作者
宗影
李玉凤
刘红玉
Zong Ying;Li Yufeng;Liu Hongyu(School of Marine Science and Engineering,Nanjing Normal University,Nanjing 210023,China)
出处
《南京师范大学学报（工程技术版）》
CAS
2021年第4期47-55,共9页
Journal of Nanjing Normal University(Engineering and Technology Edition)
基金
国家自然科学基金项目(31971547、41871188)
江苏省科技厅重点研发项目(BE2018681).
关键词
植被分类
面向对象
随机森林
特征选择
滨海湿地
vegetation classification
object-oriented
random forest
feature selection
coastal wetland
作者简介
通讯作者:刘红玉,博士,教授,研究方向:景观生态学和滨海湿地景观生态,E-mail:liuhongyu@njnu.edu.cn。
