River-Net:面向河道提取的Refined-Lee Kernel深度神经网络模型  

River-Net:A Novel Neural Network Model for Extracting River Channel Based on Refined-Lee Kernel

作　　者：李宁[1,2,3] 郭志顺 毋琳 赵建辉[1,2,3] LI Ning;GUO Zhishun;WU Lin;ZHAO Jianhui(College of Computer and Information Engineering,Henan University,Kaifeng 475004,China;Henan Engineering Research Center of Intelligent Technology and Application,Kaifeng 475004,China;Henan Key Laboratory of Big Data Analysis and Processing,Kaifeng 475004,China)

机构地区：[1]河南大学计算机与信息工程学院,开封475004 [2]河南省智能技术与应用工程技术研究中心,开封475004 [3]河南省大数据分析与处理重点实验室,开封475004

出　　处：《雷达学报（中英文）》2022年第3期334-344,共11页Journal of Radars

基　　金：河南省重大公益项目(201300311300);河南省科技攻关计划项目(212102210101,212102210093);国家自然科学基金(61871175)。

摘　　要：高精度提取合成孔径雷达(SAR)图像中的河流边界,对河流水势监测具有重要意义。以检测郑州7·20暴雨后黄河的健康状况为实施例,该文融合精致Lee滤波思想与卷积操作的滤波特性,提出了基于河道几何特性的优化内部权值卷积核Refined-Lee Kernel,进而提出了一种新型河道提取深度神经网络模型,即River-Net。为验证所提模型的有效性,该文获取了郑州7·20暴雨前后两景欧空局Sentinel-1卫星20 m分辨率干涉宽幅(IW)影像数据,利用暴雨前的影像对模型进行训练,用于提取暴雨后的黄河河道,分析黄河在暴雨后的涨势情况。实验结果表明,相比主流语义分割模型,所提模型能够更精确地在SAR图像中提取河道,对洪水灾害的检测与评估有重要应用价值。High-precision extraction of river boundaries in Synthetic Aperture Radar(SAR)images is of great significance in monitoring rivers.In this paper,the detection of the health of the Yellow River after the rainstorm in 20 July,2021 in Zhengzhou is the focus of this paper.The refined-Lee filtering concept and the filtering characteristics of the convolution operation are combined,and an optimized internal weight convolution kernel Refined-Lee Kernel is proposed according to the geometric characteristics of the river channel.A novel river extraction deep neural network model,the River-Net,is also proposed.To verify the effectiveness of the proposed model,this article utilized 20 m resolution Interferometric Wideswath(IW)image data obtained from the European Space Agency Sentinel-1 satellite before and after the 20 July rainstorm in Zhengzhou,employing the images before the rainstorm to train the model.The model,after training,was used to extract the Yellow River channel and analyze the rise of the river after the rainstorm.Experimental results show that the proposed model can extract river channels from SAR images more accurately than trendy semantic segmentation models.The model has important application value for flood disaster detection and evaluation.

关 键 词：合成孔径雷达(SAR) Refined-Lee Kernel 精致Lee滤波 神经网络 河道提取 

分 类 号：TN959.1[电子电信—信号与信息处理] TP183[电子电信—信息与通信工程]