【摘要】 为提高农田洪涝灾害信息提取效率，探索了有效的遥感影像水体自动提取方法，以2021年7月下旬河南省浚县农田洪涝灾害为研究对象，在灾前、灾中和灾后以Sentinel-2遥感数据归一化水体指数（NDWI）、改进的归一化水体指数（MNDWI）、多波段水体指数（MBWI）和B12波段作为多维特征，采用多维非监督水体自动提取方法提取水体面积。同时利用Canny-Edge-Otsu水体自动提取方法分别对MBWI、MNDWI、NDWI 3种水体指数和GF-3数据的HV极化波段进行阈值分割，提取农田洪涝灾害信息，并对不同数据源和不同方法提取农田洪涝灾害信息的精度进行对比分析。结果表明，多维非监督水体自动提取方法集成了多种水体指数和波段作为多维特征，灾前和灾后水体提取误差分别为6.99%和7.45%，低于Canny-Edge-Otsu水体自动提取方法；MBWI、MNDWI与NDWI 3种水体指数相比，MBWI水体提取误差最小，NDWI提取误差最大，但均易将建筑物和云阴影地区误判为水体；灾后基于GF-3的洪水提取误差为15.57%，高于Sentinel-2影像，但GF-3遥感影像不受云雨天气影响，能够在洪涝灾害应急监测中提供有力的数据支撑。更多还原

【Abstract】 In order to improve the efficiency of information extraction of farmland flood disaster,an effective method of automatic extraction of water body from remote sensing images was explored. The flood disaster in Xunxian County,Henan Province in late July 2021 was taken as the research object,and the NDWI(normalized difference water index),MNDWI(modified normalized difference water index),MBWI(multi-band water index)and B12 band of Sentinel-2 remote sensing data were taken as the multi-dimensional characteristics before,during and after the disaster,and the multi-dimensional unsupervised water body automatic extraction method was used to extract the water body area. At the same time,Canny-Edge-Otsu automatic water body extraction method was used to segment MBWI,MNDWI,NDWI and the HV polarization band of GF-3 data respectively to extract farmland flood disaster information,and the accuracies of different data sources and methods to extract farmland flood disaster information were compared and analyzed. The results showed that the multi-dimensional unsupervised water body automatic extraction method integrated various water body indexes and bands as multi-dimensional features,and the extraction errors of pre-disaster and disaster water bodies were6. 99% and 7. 45% respectively,which were lower than those of Canny-Edge-Otsu automatic water body extraction method. By comparing NDWI,MBWI and MNDWI,MBWI had the smallest extraction error and NDWI had the largest extraction error,but buildings and cloud shadow areas were easily mistaken for water bodies. The error of flood extraction based on GF-3 after the disaster was 15. 57%,which was larger than Sentinel-2 image. However,GF-3 remote sensing image was not affected by cloud and rain weather,so it provided a strong data support in emergency monitoring of flood disaster.更多还原