Coastline Contour Extraction Approach Research Based On GF-3 SAR Remote Sensing Image

Synthetic Aperture Radar(SAR)is an active sensor using microwave imaging,which improves the range resolution by pulse compression technology and the azimuth resolution by synthetic aperture technology.It is widely applied in military reconnaissance,resource exploration,geological surveying and mapping,marine monitoring and other fields.Compared with the optical remote sensing,microwave imaging has the advantage that cannot be influnced by climatic condition,and can also obtain the ground data even in cloudy or rainy weather conditions,which is very important for coastal information investigation.However,the speckle noise that produced in the process of SAR imaging makes it difficult in separating the target and background of sea.As a result,coastline extraction cannot be effectively completed in many cases and the application of SAR data was limited,so it is very necessary to study the coastline detection and extraction approach aiming at SAR image data.In this paper,according to the SAR remote sensing image obtained by the first C band GF-3 satellite of our country as well as the RADARSAT-2 satellite,we realized various methods to extract the coastline contours of the target sea area in the experiment.By the comparison with optical remote sensing images of the same sea area and analyzed the results of visual interpreting,we found that Regional Distance Regularized Geometric Active Contour Models(RDRGAC)has a better result in the experiment.And we establish a nonlinear fitting relationship between the area energy coefficient ? and Equivalent Number of Looks(ENL)based on different levels of noise image,which adjust the different content of SAR image noise.it is obviously improved the original methods in the coastal Contour extraction accuracy of graphics and iteration speed.In this paper,RDRGAC method is put forward in the research and experiment,and we achieved the coastline extraction of Beidaihe in Hebei province,Jinzhou bay in Dalian,local area of Tianjin port,and get the following conclusion:1.Traditional methods such as threshold segmentation,edge detection algorithms is susceptible to noise interference and the contour of coastline extraction is discontinuity when used in SAR image.Also,coastline extraction process stay at the grid level,not directly to vector coastline boundary correction,which increase the computing efforts in actually work.Parameter active contour model GVF is sensitive to noise,and can't handle contour line splitting combined,easy to fall into local minimum value.The energy function of traditional geometric active contour model CV model is a convex function and segmentation result of level set function is sensitive to the initial setting,which cannot converge to coastline for the L1 raw SAR data distributed with uneven strong speckle noise.2.The experimental results show that the RDRGAC model has a better performance than traditional active contour models and original parameters method.For the strong edge weak noise image,fuzzy edge of weak noise image,strong edge of strong noise image,this method both can realize the adaptive adjustment according to different noise levels and improve the coastline automatic extraction accuracy.It is proved that the parameter ? adaptive range is 1 ?? ?5,which has stronger robustness in heterogeneous areas of gray scale difference on SAR image,as well as inhibition for speckle noise.The improved parameters of the RDRGAC model is well adapted for different noise content of RADARSAT-2 and GF-3 remote sensing images.