基于Sentinel-2影像红边光谱指数与特征优选的竹林提取研究  

Bamboo Information Extraction from Sentinel-2 Image Based on Improved Spectral Indices and Random Forest Gini Index

作　　者：姚茂林 江洪[1] 张丽玉 Yao Maolin;Jiang Hong;Zhang Liyu(Key Lab of Spatial Data Mining and Information Sharing of Ministry of Education,National&Local Joint Engineering Research Center of Satellite Geospatial Information Technology,Academy of Digital China(Fujian),Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学空间数据挖掘和信息共享教育部重点实验室,卫星空间信息技术综合应用国家地方联合工程研究中心,数字中国研究院(福建),福建福州350108

出　　处：《海南大学学报（自然科学版）》2022年第4期373-381,共9页Natural Science Journal of Hainan University

基　　金：福建省科技计划引导性项目(2021Y0005)。

摘　　要：为了从Sentinel-2A影像中快速、准确提取竹林分布信息,以福建省永安市上坪乡竹林为研究区开展竹林提取研究.在影像分割的基础上,提取原始波段光谱、红边光谱指数、纹理3类共18个特征变量,利用随机森林Gini系数法进行特征变量重要性排序,设计5种不同特征变量组合方案,采用随机森林分类进行竹林分布信息提取.结果表明:原始波段光谱特征在Sentinel-2A影像竹林信息提取中具有重要作用,红边光谱指数特征次之,纹理特征未发挥显著作用.在红边光谱指数特征中,基于红边综合效应指数(MVIred1)构建的红边竹林指数3(BImvired1)具有良好的分类性能;利用随机森林Gini指标结合OOB泛化误差法有效减少了噪声数据的影响,筛选出最有利于竹林提取的特征变量子集,基于该特征子集的竹林分类总体精度(OA)达到94.58%、Kappa系数0.91、生产者精度(PA)为95.09%、用户精度(UA)85.54%.In the report,in order to rapidly and accurately extract bamboo distribution information from Sentinel-2A image,the bamboo forest in Shang ping township,Yong′an City,Fujian Province,were selected as the research area.Based on the image segmentation,18 characteristic variables,belong to original band spectrum,red edge spectrum indices,and texture characteristic,respectively,were extracted,the Random Forest Gini method was used for the characteristic variable selection,the five-combination scheme of the characteristic variables were designed,and the Random Forest classification method was used to extract bamboo forest information.The results showed that the original band spectral feature plays an important role in bamboo forest extraction from Sentinel-2A image,and the following is the red edge spectral indices,however,texture characteristic has no effects.The Red Edge Bamboo Index(BImvired1)based on Red Edge Comprehensive Effect Index(MVIred1)has good classification performance.The Random Forest Gini and OOB generalization error methods decreased the effects of noisy data effectively,and the subset of characteristic variables were obtained.Based on the subset,the bamboo forest was extracted accurately with the overall accuracy(OA)94.58%,kappa coefficient 0.91,producer accuracy(PA)95.09%and user accuracy(UA)85.54%,respectively.

关 键 词：竹林 光谱指数 随机森林Gini指标 红边综合效应指数 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]