Research On Water Body Information Extraction Algorithm And Application Based On Satellite-borne SAR Images

Synthetic Aperture Radar(SAR)is a type of active microwave sensor that can provide thorough surface observation with the all-weather and day-night applicability.Due to the unique characteristics,SAR is widely used in various applications,including earth observation and military surveillance.As SAR systems continue to evolve,the high-resolution SAR images are becoming increasingly massive.This poses a significant challenge in extracting the desired target information from these images,accurately and efficiently.In order to achieve automatic interpretation of SAR images,this paper focuses on shoreline detection and water extraction.And three improved methods are proposed in this paper to carry out the research of water information extraction in SAR images.The research and applications carried out in this paper include the following three aspects.1.A shoreline detection algorithm based on a modified Region-Scalable Fitting(RSF)model is proposed.The concept of coarse-to-fine detection is introduced in the shoreline detection method.A coarse shoreline is obtained with the OTSU algorithm,which is set as the initial shoreline in the modified RSF model.The key point of the modified RSF model is to introduce a novel edge energy term and the global energy term in the modified RSF energy function.Based on Laplacian of Gaussian operator and ratio of exponentially weighted averages operator,a novel edge energy term is constructed to accurately locate the boundary and reduce false boundary.Additionally,the global energy term is adopted to fit the global information well.The experimental results based on Sentinel-1 SAR images demonstrate that the proposed approach has a stronger ability to maintain weak edges compared with RSF model,which exhibits better effectiveness and reliability.2.A flood detection method based on the Multi-Scale Deeplab(MS-Deeplab)model is proposed.The model can make full use of the dual-polarization information and multi-scale features of SAR images.Firstly,the dual-channel feature extraction backbone based on the lightweight Mobile Net V2 separately trains the dual-polarization SAR images.And the obtained training parameters are merged to fuse dual-polarization water features.Then,a multi-scale feature fusion module is introduced to effectively utilize multi-layer features and contextual information.Finally,a joint loss function is constructed based on cross-entropy and a dice coefficient to deal with the imbalanced categorical distribution in the training dataset.The experimental results on the time series of Sentinel-1 SAR images show that the proposed method for flood detection has a strong ability to locate water boundaries and tiny water bodies,which provides powerful data support for disaster assessment.3.An urban water detection method based on the Hybrid Attention Unet(HA-Unet)model is proposed.In this method,Resnet50 is adopted as the backbone of HA-Unet to extract multi-level features of SAR images.During the feature extraction process,CSAM based on local attention is adopted to enhance the meaningful water features and ignore unnecessary features adaptively in shallow layers.In deep layers,MSAB based on multi-head self-attention is introduced to capture the global attention of SAR images.The experimental results based on Sentinel-1 SAR images demonstrate show that the proposed urban water extraction method has a strong ability to extract water bodies in the complex urban areas.