Research On Water Body Extraction Method And Applications Based On Optical And SAR Data

Water is a crucial resource that underpins the sustainable development of economies,societies,and ecosystems.The ability to quickly and accurately obtain information about the distribution of surface water resources is essential to address the problems of emergency monitoring represented by flood disaster monitoring and routine monitoring represented by large-scale water resources monitoring.As the availability of multi-source satellite data continues to increase,the use of remote sensing technology for water resources monitoring has garnered significant attention.This article addresses the demand for flood distribution monitoring and large-scale water resource monitoring and develops two novel water body extraction methods using synthetic aperture radar(SAR)images and optical images,respectively.The methods are applied to conduct flood disaster monitoring in Xun County,Henan Province and long-term monitoring of Poyang Lake.The main research contents and achievements are as follows:(1)Combining SAR imagery and optical imagery for water extraction method and its application in flood monitoringThe accuracy of water body extraction from SAR images is susceptible to radar shadow and speckle noise,while optical images can be restricted by cloud and fog interference.To address these challenges,this thesis proposes a novel water body index algorithm(SOWI)that combines SAR and optical images.Extensive experimental results from five different regions in China under varying weather conditions demonstrate the SOWI algorithm’s ability to achieve high-precision and automated water body extraction in diverse terrain and landform features.The SOWI algorithm is applied to the 2021 flood monitoring in Xun County,Henan Province,allowing for the extraction of water body distribution after the flood outbreak.Our analysis,combining the 2021 land classification data from the European Space Agency,reveals the distribution range of water bodies post-flood,and we analyze the change law of the flood submersion range and the disaster situation of different land cover types.Our results provide effective reference data for flood disaster monitoring,post-disaster rescue and relief,and post-disaster reconstruction.(2)Water extraction method based on deep learning for Sentinel-2 imagery and its application in lake long time series monitoringDeep learning models can achieve large-scale,long-term,and high-precision water body extraction by learning spectral and textural features of water bodies in remote sensing images.Currently,there is no publicly available water body labeling dataset specifically designed for Sentinel-2 with a resolution of 10 meters.This study develops a water body sample dataset using automated extraction and manual visual interpretation methods.The deep learning models are trained on this dataset and compared with traditional unsupervised classification models.Results show that the deep learning model trained on the water body sample dataset developed in this study has significant advantages in extraction accuracy,versatility,and computational efficiency.Based on Sentinel-2 satellite image data,this model is used to obtain the water body distribution map of Poyang Lake from 2016 to 2022 and analyze the changing pattern of water bodies.The study identifies the areas of Poyang Lake that are vulnerable to flooding during the flood season.Combining the width data of the rivers that flow into Poyang Lake,the study analyzes the rivers that are most closely related to changes in Poyang Lake area.This provides scientific reference information for flood prevention during the flood season,drought disaster control,and ecological water replenishment of Poyang Lake.(3)Design and implementation of remote sensing image water automatic extraction and application softwareIn response to the demand for automated water extraction,flood monitoring,and long-term lake monitoring based on remote sensing imagery,this thesis designed a software with a visual interface using the Py Qt library in the Python 3.7 environment.The software implements various modules such as remote sensing image preprocessing,flood disaster monitoring,and long-term lake monitoring using highperformance libraries including GDAL,Py Torch,Num Py,and Open CV.