基于中高分辨率遥感影像的水体提取研究进展  

Advances in water body extraction from medium-to-high resolution remote sensing imagery introduction

作　　者：刘憬豫 宋开山[1] 刘阁 周亚明 王玉 Liu Jingyu;Song Kaishan;Liu Ge;Zhou Yaming;Wang Yu(Northeast Institute of Geography and Agroecology,Chinese Academy of Sciences,Changchun 130102,Jilin,P.R.China;State Environmental Protection Key Lab of Satellite Remote Sensing,Ministry of Ecology and Environment Center for Satellite Application on Ecology and Environment,Beijing 100094,P.R.China;Jilin Agricultural University,College of Information Technology,Changchun 130118,Jilin,P.R.China)

机构地区：[1]中国科学院东北地理与农业生态研究所,吉林长春130102 [2]生态环境部卫星环境应用中心国家环境保护卫星遥感重点实验室,北京100094 [3]吉林农业大学信息技术学院,吉林长春130118

出　　处：《湿地科学》2025年第1期72-83,共12页Wetland Science

基　　金：国家重点研发计划项目(2024YFD1500602,2022YFB3903502);国家自然科学基金项目(42171385)资助。

摘　　要：水是地球上最重要的自然资源之一,对于地球上任何生物的生存及发展都有着至关重要的作用。利用遥感数据进行水体监测有助于环境监测、水资源管理、农业及工业生产,其中水体提取是水资源监测的重要前提。本文基于中高分辨率卫星遥感影像对水体提取的方法进行了全面综述,从文献计量学角度分析当前水体提取研究的热点,从数据源和提取方法两方面展开综述。在数据源方面,对光学遥感影像、雷达遥感影像和无人机影像进行优缺点分析,其中,光学遥感影像覆盖范围广、时间序列长,适用于大尺度及长时间序列分析的水体监测;雷达遥感影像不受云雾等影响,可实现全天候监测,适用于对天气条件较为敏感的水体监测;无人机使用灵活,分辨率高,适用于灾害评估或需进行水体边界信息精细化提取的水体监测。在提取方法方面,对阈值分割法、机器学习法和深度学习法进行比较,结果表明,目前基于深度学习方法的中高分辨率影像水体提取的精度较高,基本可达到90%以上。最后针对当前研究的不足做出总结,强调了不同方法的适用场景和局限性,并提出了未来研究的展望和建议。本文对于提高中高分辨率影像水体提取的效率和精度,促进水资源管理和环境监测等应用领域的发展具有重要意义。Water is one of the most critical natural resources on the Earth,playing a crucial role in the survival and development of all living organisms.Utilizing remote sensing data for water body monitoring is helpful for environmental surveillance,water resource management,agricultural food production,and industrial activities,with water body extraction serving as a crucial prerequisite for effective water resource monitoring.This paper provides a comprehensive review of water body extraction methods based on medium-to-high resolution satellite imagery,analyzing current research hotspots from a bibliometric perspective and examining both data sources and extraction methodologies.In terms of data sources,optical remote sensing imagery offers broad coverage and long time series,making it suitable for large-scale and long-term water body monitoring despite limitations due to cloud cover;radar remote sensing imagery enables all-weather monitoring unaffected by clouds or fog,ideal for areas sensitive to weather conditions;unmanned aerial vehicle(UAV)imagery provides high flexibility and resolution,particularly useful for disaster assessment and detailed extraction of water body boundaries,but may face regulatory and operational challenges.Regarding extraction methods,threshold segmentation methods utilize simple thresholding based on spectral indices like NDWI or MNDWI for computational efficiency,though with limited accuracy in complex environments;machine learning methods such as Support Vector Machines(SVM)offer improved precision by classifying pixels into water and non-water categories;deep learning methods achieve high accuracy rates exceeding 90%for water body extraction from medium-to-high resolution imagery,demonstrating superior performance.Finally,this review summarizes the shortcomings of current research,emphasizing the appropriate scenarios for each method and acknowledging their limitations while proposing future research directions and recommendations.These findings are crucial for advancing water resource m

关 键 词：水体识别 中高分辨率卫星遥感 遥感监测 机器学习 神经网络 

分 类 号：X87[环境科学与工程—环境工程]