1、引用格式：刘文宋,张仲英,郑琳,等.基于改进HLT与深度学习的双时相PolSAR洪涝灾害监测新方法J.地球信息科学学报,2023,25(8):1730-1745.Liu W S,Zhang Z Y,Zheng L,et al.A novel flood disaster monitoring method based on improved HLT and deep learning usingbi-temporal PolSAR imagesJ.Journal of Geo-information Science,2023,25(8):1730-1745.DOI:10.12082/dqxxk
2、x.2023.220985基于改进HLT与深度学习的双时相PolSAR洪涝灾害监测新方法刘文宋，张仲英，郑琳，郭风成*江苏师范大学地理测绘与城乡规划学院，徐州 221116A Novel Flood Disaster Monitoring Method based on Improved HLT and Deep LearningUsing Bi-temporal PolSAR ImagesLIU Wensong,ZHANG Zhongying,ZHENG Lin,GUO Fengcheng*School of Geography,Geomatics and Planning,Jiangsu
3、Normal University,Xuzhou 221116,ChinaAbstract:Rapidly and accurately monitor the flood disaster is very important,which can protect peoples lifeand property safety and realize the sustainable development of society.The polarimetric synthetic aperture radar(PolSAR)image can obtain all-day and all-wea
4、ther information of flood disaster by transmitting and receivingelectromagnetic microwave of different polarizations,which can provide more favorable data support for flooddisaster monitoring.However,the traditional flood disaster monitoring methods based on PolSAR images areseriously affected by sp
5、eckle noise and the class imbalance between changed class and un-changed class leads tolow accuracy of disaster monitoring.To address issues,a novel flood disaster monitoring method based on theimproved Hotelling-Lawley Trace(HLT)statistic operator and deep learning for small area change using bi-te
6、mporal PolSAR images was proposed in this paper.Within this method,the HLT statistic operator was firstlyconstructed by considering the neighborhood information of PolSAR images,which can reduce the influence ofspeckle noise and spatial heterogeneity for the generation of the difference image in thi
7、s paper.Secondly,theTwo-stage Center-Constrained Fuzzy C-Means clustering(TCCFCM)algorithm and the Deep ConvolutionalGenerative Adversarial Network(DCGAN)were introduced to build a robust method of sample select andexpress when lacked the changed samples over disaster areas,which can solve the probl
8、em of the classimbalance between changed and un-changed class.Finally,the Deep Convolutional Wavelet Neural Network(DCWNN)was constructed to achieve accurate monitoring of flood disasters.In order to verify the feasibilityand robustness of the proposed method,the Radarsat-2 images covered Wuhan city
9、 before and after flood收稿日期：2022-12-19；修回日期：2023-01-06.基金项目：国家自然科学基金项目（62201232、62101219）；江苏省自然科学基金项目（BK20210921、BK20201026）。Foundation items:National Natural Science Foundation of China,No.62201232,62101219;Natural ScienceFoundation of JiangSu Province,No.BK20201026,BK20210921.作者简介：刘文宋（1988），男，山东潍坊
10、人，博士，讲师，主要从事PolSAR影像解译方法研究。E-mail:*通讯作者：郭风成（1992），男，江苏宿迁人，博士，副教授，主要从事SAR影像增强研究。Email:Vol.25,No.8Aug.,2023第25卷 第8期2023年8月8期刘文宋 等：基于改进HLT与深度学习的双时相PolSAR洪涝灾害监测新方法disaster were selected in July,2016.The qualitative and quantitative results show that the proposed method canreduce the missed alarm rate an
11、d false alarm rate of flood disaster monitoring,and significantly improve theOverall Accuracy(OA)and Kappa coefficient(Kappa)of disaster monitoring,i.e.the false alarm rate andmissed alarm rate of flood disaster monitoring are reduced respectively 1.5%and 2%,meanwhile,the OA andKappa coefficient of
12、flood disaster monitoring are significantly improved 3%and 0.02 when compared with thetraditional methods of flood disaster monitoring,which can provide technical support of emergency relief workfor relevant departments.Key words:flood disaster;PolSAR image;improved HLT operator;class imbalance;TCCF
13、CM algorithm;DCGAN;DCWNN*Corresponding author:GUO Fengcheng,E-mail:摘要：全极化合成孔径雷达（PolSAR）通过主动收发不同极化方式的微波信号，可为全天时、全天候获取洪涝受灾信息提供有利的数据支持。然而，传统基于PolSAR影像的洪涝灾害监测方法受相干斑噪声影响严重，且洪涝灾害引发的变化类与非变化类的类别不平衡易导致灾害监测精度低。针对以上问题，本文提出了一种基于改进HLT与深度学习的双时相PolSAR洪涝灾害监测新方法。首先，通过构建一种顾及邻域信息的改进Hotelling-Lawley迹（HLT）统计量算子，减少PolSAR
14、影像相干斑噪声及空间异质性对差异影像生成的影响；其次，为解决洪涝受灾区域变化类样本不足及变化类与非变化类不平衡的问题，引入双阶段中心约束FCM（TCCFCM）算法与深度卷积对抗生成网络（DCGAN）模型，形成一种稳健的洪涝灾害样本选择与扩充方法；最后，通过构建一种深度卷积小波神经网络（DCWNN）模型实现洪涝灾害精确监测。为了验证本文方法的可行性与鲁棒性，本文选取了2016年7月武汉梁子湖与严东湖洪涝灾害发生前后Radarsat-2影像进行实验。定性与定量评价结果表明：相对于传统的洪涝灾害监测方法，本文方法综合降低洪涝灾害监测的虚警率与漏警率分别为2%及1.5%左右，而监测洪涝灾害的总体精度与
15、Kappa系数可分别提升3%及0.02左右，为相关部门应急救灾等工作提供技术支持。关键词：洪涝灾害；PolSAR影像；改进HLT算子；类别不平衡；双阶段中心约束FCM算法；深度卷积对抗生成网络；深度卷积小波神经网络1 引言近年来，我国洪涝灾害多地频发，如2016年武汉、2018年广东及2021年郑州洪涝灾害等，严重损害了当地人民生命财产安全，阻碍了经济、社会可持续发展1-2。及时、准确地监测洪涝灾害也是促进社会高质量发展的重要保障之一3。相对于传统洪涝灾害实地调查手段，基于遥感影像的洪涝灾害监测技术可以快速、大范围监测洪涝灾害的影响，为助力相关部门防灾减灾提供技术支持4。然而，在洪涝灾害发生时
16、，常伴随阴雨天气，传统光学遥感传感器由于受工作波段较短、且为被动成像等条件限制，难以及时、快速获取洪涝受灾区域的高质量影像，影响后续相关应急救援工作实施5-6。合成孔径雷达（Synthetic Aperture Radar，SAR）通过主动微波成像方式，可全天时、全天候获取不受云雾影响的灾区影像7-8。此外，相对于单极化SAR影像，全极 化 SAR（Polarimetric Synthetic Aperture Radar，PolSAR）通过收发不同极化方式的电磁波，可以获取更加丰富的受灾地物的散射特性，为精准洪涝灾害监测提供可靠的数据支撑9。因此，基于不同时相PolSAR影像的洪涝灾害监测方法已成为近年该领域研究的热点。传统基于不同时相 PolSAR 影像的洪涝灾害监测方法一般包括：影像预处理、差异影像生成及差异影像分析3个过程7,10。其中，差异影像生成与差异影像分析是洪涝灾害监测的核心11。近年来，国内外专家学者提出了众多基于PolSAR影像的差异影像生成算法，如对数比值法、主成分分析法、非局部图法等12，但以上方法大多只针对Pol-SAR 影像的强度或幅度信息，尚未充分利用 P
