基于多时相Sentinel-2数据的温州市红树林提取及应用  

Extraction of Mangroves in Wenzhou City Based on Multi-temporal Sentinel-2 Data and Its Application

作　　者：廖孟光[1,2,3] 陈醒 李猛 李朝奎 王韫[1,2,3] LIAO Mengguang;CHEN Xing;LI Meng;LI Chaokui;WANG Yun(Sanya Institute,Hunan University of Science and Technology,Sanya 572024,China;School of Earth Sciences and Spatial Information Engineering,Hunan University of Science and Technology,Xiangtan 411201,China;National-Local Joint Engineering Laboratory of Geo-Spatial Information Technology,Hunan University of Science and Technology,Xiangtan 411201,China)

机构地区：[1]湖南科技大学三亚研究院,海南三亚572024 [2]湖南科技大学地球科学与空间信息工程学院,湖南湘潭411201 [3]湖南科技大学地理空间信息技术国家地方联合工程实验室,湖南湘潭411201

出　　处：《湖南科技大学学报(自然科学版)》2024年第4期37-45,共9页Journal of Hunan University of Science And Technology:Natural Science Edition

基　　金：湖南省教育厅科研项目资助(22C0254);大学生创新创业训练计划资助项目(S202310534030)。

摘　　要：为准确了解红树林在温州市的分布,更好地对该市红树林资源进行调查、维护和利用,选取温州市沿海4个区域为研究区,利用2021年—2022年的Sentinel-2多光谱影像,通过提取研究区内红树林与互花米草的归一化植被指数(NDVI)构建植被指数时序特征,分别采用随机森林(RF)、支持向量机(SVM)和最大似然法分别对单一时相和加入植被指数时序特征的多时相影像进行红树林识别.结果表明:基于多时相影像的红树林识别精度要高于单一时相影像;相比单一时相,当Kappa系数提高0.1以上时,NDVI时间序列能较好地描述红树林的物候特征和提高红树林的提取精度;3种算法均能准确地识别红树林,且随机森林算法的识别精度最高,其总体精度为98.02%,Kappa系数为0.84;研究区共识别红树林1.5269 km2,准确率达87.26%.研究结果可为温州市红树林的管理和维护提供可靠的数据基础.An Accurate understanding of the distribution of mangroves in Wenzhou is helpful for the investigation,maintenance and utilization of mangrove resources in the city.In this paper,four coastal areas of Wenzhou City are selected as the study area,and a total of five Sentinel-2 multispectral images from 2021 to 2022 are used to construct the time series characteristics of vegetation index by extracting the normalized vegetation index of mangroves and Spartina alterniflora in the study area.Random forest(RF),support vector machine(SVM)and maximum likelihood method are used to identify mangroves in single phase and multi-phase with vegetation index time series features.Results show that the recognition accuracy of mangroves based on multi-temporal images is higher than that of single-phase images.Compared with single-phase images,the Kappa coefficient is increased by more than 0.1.The study verifies that NDVI time series can better describe the phenological characteristics of mangroves and improve the extraction accuracy of mangroves;all three extraction algorithms can accurately identify mangroves,and the random forest method has the highest recognition accuracy with an overall accuracy of 98.02%and a Kappa coefficient of 0.84,with the identified mangroves in study area of 1.5269 km~2,and the accuracy of 87.26%,aiming to provide basic data for the management and maintenance of mangroves in Wenzhou.

关 键 词：红树林 Sentinel-2影像 归一化植被指数 多时相 

分 类 号：P237[天文地球—摄影测量与遥感]