| Chaohu Lake is the fifth largest fresh water lake in China.With the rapid development of economy,the land use structure in this area has also undergone great changes.As one of the pollution control projects of "Three Rivers and Three Lakes" in China,Chaohu Basin has received widespread attention,and related researches involve all directions in the basin.The information provided by multi-source remote sensing image data has the characteristics of redundancy,complementarity and cooperation,which is suitable for long-term remote sensing image water extraction,and becomes an important means for long-term monitoring of lake water changes.Traditional visual interpretation of water bodies based on optical remote sensing images has many shortcomings in terms of production cost,update cycle and timeliness.The pixel based classification method(SVM method,decision tree method)and the object-oriented method are difficult to meet the requirements of large-scale and multi-temporal water remote sensing information monitoring due to the fragmentation of classification results,complex development of object rule set and low generalization.However,due to the limitation of water index construction in visible and near-infrared bands,high resolution satellite images are difficult to achieve accurate extraction of water.Aiming at the existing problems of water extraction from remote sensing images,this paper proposes a water extraction model based on deep learning.This model introduces dense connection blocks and local feature compression modules,which can overcome the weak generalization ability of current deep learning water extraction models,enhances the model’s ability to extract image features and feature fusion capabilities,and achieves high-precision,multi-sensor water extraction tasks.This article selects Chaohu basin as study area,the proposed method was applied to water extraction of the multisource remote sensing image which can provide data support for the change detection of time and space for water and drainage pattern.The specific results are as follows:(1)Through experimental verification,the method presented in this paper has a good performance in water extraction of Gaofen-2 remote sensing image,and the best band combination of Gaofen-2 remote sensing image for water extraction is obtained which is the green,red and near-red band.The extraction of large water bodies from remote sensing images is complete,and the problems of wrong shadow and missing small water bodies are improved.Through quantitative accuracy evaluation,the extraction accuracy of the proposed method is 99.21%,97.51% and 96.50% in OA,F1-score and Io U indexes,respectively.The proposed method also has good performance in three remote sensing images of water extraction,namely Gaofen 6,Sentinel 2 and Ziyuan 3.Based on quantitative accuracy evaluation,the extraction accuracy of the proposed method in the Gaofen-6 image is 99.42%,99.2% and 98.42% in OA,F1-score and Io U indexes,respectively.Therefore,the proposed method realizes the rapid and automatic extraction of water bodies from multi-source high-resolution optical remote sensing images.It can quickly and accurately extract water information,which can provide technical support and data support for relevant decision-making departments.(2)The water data obtained from different data sources were studied in different aspects.1.Define the degree of river bending and fracture,and perform quantitative calculations on the degree of river bendingBased on the water data obtained from Gaofen-2 image,we proposed an index to evaluate the sinuous degree and analyzed the meandering degree of the river in quantitatively in combination with the river bending.2.Flood disaster monitoringAccording to the extraction results of Gaofen-6 water body,Chaohu Lake Basin was seriously affected by flood in 2020.After the flood disaster,the body water body of Chaohu Lake became larger,the width of the river channels in the basin increased,and five large-scale submerged areas appeared in the Chaohu Lake Basin.3.Long-term lake area change monitoring and water system spatial distribution characteristicsThe water body data obtained from Landsat 8 images during 2013-2018 were used to obtain the change rule of water body area in 6 years.In addition,the water extraction results in 2013 were further processed to obtain the water system distribution map of Chaohu Lake Basin,and the water system produced by two DEM with different resolutions was compared and analyzed.The water system data obtained by the method in this paper were used to analyze the spatial characteristics of the water system pattern.In general,from 2013 to 2018,the water area of Chaohu Lake Basin showed a basically stable trend,and there was a small increase.In 2013,Wuwei County,Chaohu City,Chaohu City,Lujiang County,Wuwei County,and Feixi County had the largest firstorder average branching ratio of river network density,water surface ratio,river frequency and river network development coefficient in the six counties of the Chaohu Lake Basin. |
PDF Full Text Request