Flood Extent Detection In PolSAR Images Considering Prior GeoInformation

In recent years,the global climate changes has become more severely and the risk of natural disasters in China has been further increased.The frequent occurrence of nature disasters has do huge harm to the people's life and property.The extraction of water extent is an important part of the research in the area of water area cover area changes and flood disaster monitoring.The extraction of flood extent has great significance to the detection of flood disaster.Remote sensing based detection methods with satellite images obtained can be used to carry out flood extraction and analysis in a short period.So it is an efficient method for extracting flood and waterlogging disasters based on remote sensing images.Because of the serious obstruction of cloud and fog,the imaging conditions of optical satellites are limited.As a result of its active imaging mode,the imaging band of synthetic aperture radar(SAR)is a long wavelength microwave with strong penetration ability and can be imaging under a certain thickness of bad weather conditions.Therefore,the SAR satellite can provide reliable and effective image data protection for remote sensing monitoring flood disaster area.At present,we can get the scope of flood inundation after we get the flood waters or results of change detection from the post-disater image based on the pre geoinformation,such as the surface water vector data,the topology data etc.The practice shows that the detection of the extent of flood based on the heterologous remote sensing images and the extraction of SAR based on the prior geographic information are very necessary in the actual disaster reduction business.However,different with SAR sensors,the imaging mode of opitical sensors is passive,the flood extent can not be extracted preceisely through the conventional change detection process based on optical images and SAR images.Besides that,Extraction the flood with the prior vector information and SAR data and avoiding the interference of other weak scatterers are also the focus of the research.1)flood extent rapid extractions method based on improved U-Net considering the heterologous remote sensing data of optical sensors and Radar sensors.The conventional method based on heterologous remote sensing imagery would be redundant and complex.Water bodies were extracted from pre and post images,separately,then difference map was calculated based on the different extracted water bodies.Therefore,these methods would get various results because the remote sensing data could be very different and the extraction method could be sensitive to these data.In this paper,based on the basic U-Net network structure multi temporal remote sensing data,the imporved U-Net structure is proposed.The core idea is to convolution and pool several layers of optical images,and then integrate the SAR image into the deconvolution process,and finally the size of the original image is sampled.The final improved U-Net network can realize the unified and rapid extraction of the range of the optical radar special source image.2)Accurate extraction of flood extent in SAR images assisted by geographic information dataThe flood extent extracted by image can not be effectively guaranteed by multi temporal remote sensing data.And the accuracy can be further improved by combining the basic vector data.The main difficulty of geographic information assistance is how to effectively combine the boundary geometry and location information of the object water in remote sensing images,thus driving the precise evolution of the flood boundary.Therefore,the improved polarization SAR level set segmentation algorithm is first proposed in this paper.The prior water v a zero initial leve set.In this letter,a novel flood detection method based on thelevel set method and prior Geoinformation,including the topography infromation and PFPWV data,is proposed.Our main contributions are trifold.We combine the slope information derived from a DEM as an extra constraint force on the level set evolution in PolSAR images.The initialize the zero level set curves with the PFPWV data to make the boundary of flood more accurate.We propose a refinement processing method related to Kullback-Leibler divergence of circular polarization coherence.3)Post-processing and Assessment IndexBased on the theory of probability graph model,especially the Markov random field model considering neighborhood information,the results of the range extraction of flood disaster are processed after the rough boundary of the flood area and the serious noise extraction results.In view of the problem of boundary precision assessment for flood disaster extent,the edge similarity detection index(ESDI)based on shape similarity is proposed to evaluate the boundary precision of the extraction results.