Water Detection Of SAR Image Based On Improved Level Set And Spatial Structure Features

In recent years,with the deterioration of the natural eco-environment,water resources have attracted attention as a necessary factor for human survival and social development.Monitoring and management of surface water resources is not only conducive to seasonal flood monitoring,agricultural sustainable development,reasonable distribution of water resources,but also have a great significance for marine management,dam construction,and port bridge testing.Therefore,scholars from all countries are actively involved in the research of water detection methods.In the current methods,Synthetic Aperture Radar(SAR)is not affected by adverse weather,and can monitor ground objects 24 hours without interruptions,which has received great attention.However,the intensity inhomogeneity caused by the intrinsic speckle noise of SAR image,radar parameters and external environment is a major problem for the water detection of SAR image.Most of the existing water detection methods do not consider this complexity,resulting in the inability to accurately extract water body in real time.Level set is essentially a segmentation method based on edge detection,which can well respond to changes in the target topology and is widely used in SAR image segmentation.Based on the classical level set method,this paper proposes two improved level set algorithms to achieve accurate extraction of SAR water body.(1)Aiming at the intensity inhomogeneity in SAR water image,a level set segmentation method based on bias field and Gamma statistical model is proposed.The bias field is actually a quantitative description of the degree of slow change in image grays,and was first applied to medical image.However,by analyzing the SAR image,it is found that there is also an bias field in the SAR image.By embedding the bias field into the statistical level set segmentation model,it is possible to achieve segmentation of intensity inhomogeneity image.At the same time,the intensity inhomogeneity of the image can be corrected by bias field inversion.After the segmentation is completed,the features similar to the water body still exist in the water body,the spatial structure characteristics of the water body are further utilized to remove false information such as shadows and paddy fields,etc.Therefore,the accuracy of water detection is improved.(2)Aiming at SAR image with speckle noise,poor contrast and intensity inhomogeneity,a high-precision algorithm for detecting water in SAR image using multi-scale structure tensor level set model is proposed.The classical level set model mainly depends on the intensity information of the image,when the contrast of the image is extremely poor,the segmentation cannot be achieved by using the intensity information.Therefore,we introduce the structure tensor to describe the texture features of the image,and improve the classic intensity inhomogeneity image segmentation model-LBF model.First,the image texture information is added to the traditional structure tensor to smooth the texture structure.Then,the average image of all feature channel is obtained,and the global terms of the level set driving function are constructed.Finally,the LBF model is combined to obtain the level set energy function.In order to make the level set segmentation model better adapt to SAR speckle noise and reduce the time complexity,a multi-scale framework is introduced.Experiments show that the proposed method can achieve accurate water extraction of all kinds of complex SAR image with a good anti-noise performance and a low complexity.