Research On Ridge Extraction Method Based On Polarimetric SAR Image

The earthquake is one of the most abrupt and destructive nature disasters in the world.China is a country with frequent earthquakes.In recent years,the occurrence of earthquake disasters is increasing,which makes the public have more awareness of the impact of catastrophic events.Remote sensing satellites can map the area of interest quickly and with high geometric accuracy,thus providing important information about the affected areas.Since the occurrence of earthquake is often followed by bad weather conditions,the role of optical remote sensing images in disaster relief decision-making is greatly limited.However,radar remote sensing has been widely used in seismic research because of its advantages of strong penetration and not relying on sunlight.The mountain hazard triggered by earthquake usually starts near the ridgeline.Knowing the distribution of the ridgeline is conducive to making sound judgements on the possible hazards after an earthquake.However,in the complex scenes of Pol SAR images containing buildings,the ridgeline recognition process will produce incorrect judgements due to the existence of buildings.As far as available data are concerned,there are few studies on ridge recognition from Pol SAR images.Therefore,this paper firstly discusses the current research status of ridge extraction based on landmark curvature analysis and the research status of segmentation and classification of Pol SAR images by domestic and foreign scholars in recent years.Secondly,this paper introduces the basic theoretical knowledge of polarization-synthetic aperture radar image,and expounds the polarization-decomposition model and the physical scattering characteristic quantity involved in the subsequent research.Finally,aiming at the ridgeline problem in Pol SAR images,this paper takes Beichuan County,Mianyang City,Sichuan Province in 2018 as the research area,and studies the ridge extraction by using Chan-Vese model,regional growth method,variogram function and neural network model respectively,and draws the following conclusions:(1)The scattered map patches and the salt-and-pepper effect were significantly reduced when the Chan-Vese model was used to segment the image,and the polarimetric decomposition method solved the problem that the ridgeline and the building were confused together in the segmentation result.The ridgeline recognition was realized with an accuracy as high as 84.09% by using the proposed method.However,the research found that the accuracy of ridgeline recognition was only 63.58%by the traditional extraction method of Digital Elevation Model(DEM)data.The comparison between the two methods shows that the proposed method has higher recognition accuracy than the traditional methods,and can provide a new idea for ridge recognition in POLSAR images.(2)To make full use of Pol SAR images in the texture information,a novel method combined the region growing with the variogram function was proposed.This method used the automatic seeded region growing to segment the image,and then combined the variogram function with the texture entropy feature of the image.The fuzzy Cmeans method was used to extract the building areas.Finally,the ridgeline recognition result was obtained by combining the segmentation result with the extracted building areas.The results showed that the accuracy of this proposed method was 90.13%,which was 15.55% higher than that of the threshold method.(3)In order to further improve the accuracy of ridge recognition,a full convolutional neural network,pyramid attention network model,was introduced in this paper.On the basis of the polarization decomposition results,the training set and test set of the ridge and the building were obtained.Based on this,the identification of the ridges and the verification of the accuracy are realized.Experimental results show that compared with the previous two methods,the extraction accuracy of this proposed method has been greatly improved,reaching more than 98%.At the same time,this method has been greatly improved in the degree of automation,robustness and generalization ability,which is conducive to the practical application.