基于主成分变换的星载升降轨极化SAR数据水体提取方法

Water Body Extraction Method from Spaceborne Ascending and Descending Polarimetric SAR Data Based on Principal Component Analysis

利用光学遥感技术可以快速高效地获取大范围的地表水体信息,但是在洪灾监测等应用中,常常伴随多云、降雨等恶劣天气,难以及时获得高质量的光学遥感图像。合成孔径雷达(Synthetic Aperture Radar,SAR)技术具有全天时、全天候的特点,不受阴雨天气的影响,相比光学遥感具有明显的优势。本文首先总结和分析了目前基于星载SAR数据提取水体的方法和存在的问题,以Sentinel-1数据为例,为了弥补单极化阈值法和哨兵一号双极化水体指数(Sentinel-1 Dual-Polarized Water Index,SDWI)方法的不足,提出了双极化第一主成分水体指数(Dual-Polarized First-Principal-Component Water Index,DFWI)。其次,为了解决雷达阴影容易与水体相互混淆的问题,提出了3种升降轨极化SAR水体提取方法:升降轨VH极化方法(AscendingDescending VH-Polarization Water Index,AD-VH)、升降轨VV极化方法(Ascending-Descending VV-Polarization,AD-VV)和升降轨双极化第一主成分水体指数(Ascending-Descending Dual-Polarized First-Principal-Component Water Index,AD-DFWI)。最后,选择云南省洱海和土耳其Hatay 2个研究区进行水体提取实验,分别对应水体的常态化监测和应急监测等不同的应用场景。根据用户精度、生产者精度、虚警率和F1-score共4种指标对上述不同方法得到的水体提取结果进行精度评价和分析。实验结果表明:(1)本文所提方法相比单极化阈值法和SDWI法的分类精度明显提高,在2个研究区中,精度相对最高的方法均为AD-DFWI法,F1-score指标分别达到了97.83%和88.33%;(2)升降轨极化SAR水体提取方法不仅较好地解决了雷达阴影和水体相互混淆的问题,而且图像直方图中双峰分布特点更加显著,水体和非水体的可分离性更高,综合性能更好。本文提出的方法未来可以为水体提取和洪灾监测等应用提供参考。

Optical remote sensing technology can efficiently capture a wide range of surface water information.However,in challenging applications such as flood monitoring,it is often difficult to obtain high-quality optical remote sensing images,due to cloudy,rainfall,or other inclement weather conditions.Synthetic Aperture Radar(SAR)technology can provide images of both day and night and unaffected by cloud coverage,which provides obvious advantages over optical remote sensing.In this paper,we summarize the methods of surface water body extraction using SAR data,analyze the advantages and disadvantages of Sentinel-1 Dual-Polarized Water Index(SDWI),and propose a Dual-Polarized First-Principal-Component Water Index(DFWI).In addition,aiming at the problem that radar shadow is easily confused with waters in SAR images,three kinds of ascending-descending polarimetric radar water body extraction methods are proposed:Ascending-Descending VV-Polarization(AD-VV),Ascending-Descending VH-Polarization(AD-VH),and Ascending-Descending Dual-Polarized First-Principal-Component Water Index(AD-DFWI)methods.Finally,two experimental sites,Yunnan Province,China,and Hatay,Turkey,are selected for water body extraction,which corresponded to normalized monitoring and emergency monitoring of water bodies,respectively.The accuracy of the water body extraction is evaluated and analyzed using four metrics:User Accuracy,Producer Accuracy,False Alarm Rate,and F1-score.The experimental results show that:(1)Compared with the single-polarization method and the SDWI method,the classification accuracy of our proposed method is significantly improved.The AD-DFWI method achieves the highest accuracy in both experimental areas,e.g.,the F1-score index reaches 97.83%and 88.33%,respectively;(2)Our ascending-descending polarimetric SAR water extraction methods not only solve the problem of the fusion between shadows and waters but also generate a more prominent bimodal distribution histogram for improved discrimination of water and non-water.In flat areas that are not affected by radar shadows,the DFWI method can also obtain high-precision water body extraction results.This method is not only suitable for Sentinel-1 dual-polarization data but can also be theoretically extended to fully polarized data such as GF-3,ALOS-2,RADARSAT-2,and TerraSAR-X.If multiple satellite constellations can be used to obtain the ascending and descending orbit data at the same time in the future,or the revisit time of remote sensing satellites can be shortened,the accuracy of the ascending-descending polarimetric SAR water body extraction methods will be further improved.The method proposed in this paper can provide a reference for challenging applications such as surface water extraction and flood monitoring in the future.