Research On Shoreline Extraction In SAR Image Based On Improved Superpixel Segmentation Algorithm

Synthetic Aperture Radar(SAR),as an active microwave imaging system,has the advantages of all-weather,all-directional and high resolution.It has its own advantages in environmental monitoring,resource exploration,sea and land segmentation and other applications,so it has extremely high civil and military value.As the boundary between the sea and the land,the coastline is always in dynamic change under the influence of natural environment and human activities.It is of great significance to grasp the location of coastline quickly and accurately in the fields of Marine culture and sea area management,so it is particularly important to extract coastline from SAR images.In recent years,the application of superpixel segmentation algorithm can achieve excellent results in SAR image processing.Based on an improved superpixel segmentation algorithm,this thesis extracts coastlines from coastal SAR images.The main contents are as follows:(1)In view of the speckle noise and many small and slender areas in SAR images,an improved superpixel segmentation algorithm is applied to the coastline extraction method of SAR images.Firstly,the SAR image is initially segmented by an improved superpixel segmentation algorithm to form several superpixel clusters.The traditional superpixel segmentation method uses the Euclidean-style distance of CIELAB space as the distance measurement between pixel point and seed point.However,due to the particularity of SAR image,it does not contain optical information,and the direct application of traditional superpixel method to SAR image will lead to inaccurate segmentation results.In this thesis,the traditional superpixel method is improved.Standard Wishart distance is used to replace Euclidean distance in the traditional method as the distance measurement between pixel points and seed points,and the initialization method of the clustering center is optimized to avoid the drift of the clustering center in the segmentation process as much as possible,which greatly reduces the influence of speck noise on the segmentation results.(2)In order to avoid the influence of slender and tiny regions in SAR images on the segmentation results,this thesis adopts the threshold comparison method to merge the small areas in the initial segmentation results.By setting a large size threshold and a small size threshold as well as a similarity threshold,the segmented area is compared with the small size threshold.When the area is smaller than the small size threshold,the area is judged to be noise,and the area can be directly fused into the adjacent area with the highest similarity.When the region size is between the small size threshold and the large size threshold,the similarity and similarity threshold of the region and its adjacent regions will be compared.If it is larger,the region will be merged with the adjacent region with the highest similarity;otherwise,the region will be retained.(3)The texture features of the image are extracted,and the shape heterogeneity of image superpixel blocks is combined to construct the regional adjacency map.Based on the adjacency map,the segmented superpixel blocks are merged to obtain the coastline.Local Binary Patterns(LBP)can describe the structure of an image texture,while Local Contrast(LC)can describe the intensity of the texture.In this thesis,the joint probability distribution histogram of LBP and LC will be used to describe the texture features of the image.In this way,both the texture structure of the image and the texture intensity of the image will be taken into account.In addition,combining the heterogeneity of regional shape features,namely the influence of the length of the common edge of the superpixel block,the merging cost between regions will be defined,then get the shoreline.