Level set method has been extensively used for image segmentation, which is a key technology of water extraction. However, one of the problems of the level-set method is how to find the appropriate initial surface parameters, which will affect the accuracy and speed of level set evolution. Recently, the semantic segmentation based on deep learning has opened the exciting research possibilities. In addition, the Convolutional Neural Network (CNN) has shown a strong feature representation capability. Therefore, in this paper, the CNN method is used to obtain the initial SAR image segmentation map to provide deep a priori information for the zero-level set curve, which only needs to describe the general outline of the water body, rather than the accurate edges. Compared with the traditional circular and rectangular zero-level set initialization method, this method can converge to the edge of the water body faster and more precisely;it will not fall into the local minimum value and be able to obtain accurate segmentation results. The effectiveness of the proposed method is demonstrated by the experimental results of flood disaster monitoring in South China in 2020.

Chuan Xu;Shanshan Zhang;Bofei Zhao;Chang Liu;Haigang Sui;Wei Yang;Liye Mei

School of Computer Science,Hubei University of Technology,Wuhan,China;Geographic Information Center,GuangZhou UrbanPlanning&Design Survey Research Institute,Guangzhou,China;State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan,China;School of Information Science and Engineering,Wuchang Shouyi University,Wuhan,China;The Institute of Technological Sciences,Wuhan University,Wuhan,China

地球空间信息科学学报（英文版）

[1]Chuan Xu;Shanshan Zhang;Bofei Zhao;Chang Liu;Haigang Sui;Wei Yang;Liye Mei-.SAR image water extraction using the attention U-net and multi-scale level set method:flood monitoring in South China in 2020 as a test case)[J].地球空间信息科学学报（英文版）,2022(02):155-168

SAR,image,water,extraction,using,attention,net,multi,scale,level,set,method,flood,monitoring,South,China,test,case,Level,has,been,extensively,used,segmentation,which,key,technology,However,one,problems,find,appropriate,surface,parameters,will,affect,accuracy,speed,evolution,Recently,semantic,deep,learning,opened,exciting,research,possibilities,In,addition,Convolutional,Neural,Network,shown,strong,feature,representation,capability,Therefore,this,paper,obtain,map,provide,priori,information,zero,curve,only,needs,describe,general,outline,body,rather,than,accurate,edges,Compared,traditional,circular,rectangular,initialization,can,converge,faster,more,precisely,not,fall,into,local,minimum,value,able,results,effectiveness,proposed,demonstrated,by,experimental,disaster

0.576738