Research On Ship Contour Extraction Methods In SAR Images

Synthetic Aperture Radar(SAR)imaging technology has been widely applied in target recognition,ground-object monitoring etc.As one of the most important techniques of SAR image interpretation,ship contour extraction can provide the sizes,shapes and heading information of ships for ship classification and recognition,and thus enormously contributes to the acquirement of real-time marine military intelligence.In recent years,there are considerable methods for ship detection in SAR images.However,little attention is paid to ship contour extraction methods.Ship contour extraction in SAR images is often composed of ship detection in large-scene SAR images and contour extraction of each detected ship.In this thesis,we research ship contour extraction methods in large-scene SAR images based on the superpixel segmentation,region growing,target detection,and the active contour algorithm.The content of this thesis is organized as follows:In the second chapter,existing target contour extraction methods are reviewed.Firstly,segmentation methods based on region information are introduced,including the basic principle of superpixel segmentation method and the region growing model.Then,the traditional model-based curve evolution methods are introduced,including snake model,Chan-Vese model and geodesic active contours(GAC)model.The properties and limitations of the active contour model are analyzed.Finally,the flowchart based on the level set evolution model of contour extraction is introduced.In the third chapter,a ship contour growing algorithm based on superpixel segmentation and region growing is proposed for contour detection of multiple ships in SAR images.In the proposed method,superpixel segmentation based on finite mixture models(FMMs)is firstly applied to a SAR image to obtain the superpixel-level partition of the scene,and then the statistical histogram is utilized to extract superpixels with ships as the seed region,where the grayscale average value in each extracted superpixel is required to be higher than a predefined threshold.In terms of the region growing criterion,a seed region parallelly grows outwards along its edge to generate the final contour of a ship.Thanks to the parallel growing,the proposed method realizes contour detection of multiple ships inlarge-scene SAR images and improves the efficiency of ship contour extraction.By the evaluation on real SAR images,the validness of the proposed method is verified.In the fourth chapter,in order to extract contours of ships that have large difference in size in large-scene SAR images,a fast ship contour extraction method is presented by combining ship detection with active contour model.The ship slices are extracted by BING(Binarized Normed Gradients)based on supervised support vector machine(SVM)in a SAR image.Then,the morphological filtering is used to obtain the initial ship contour in each slice.Third,using the initial contour as the initial curve,the active contour model algorithm is used to iteratively generate the final contour of each ship.In comparison with existing methods on real SAR image ship data,the proposed method exhibits more effectiveness and accuracy.