Ship Detection Algorithm For Optical Remote Sensing Images Based On Sketch Information

Optical remote sensing image has attracted much attention because of its clear imaging,rich image details,intuitive and easy to understand.Object detection and recognition of optical remote sensing images is a hot research topic in the field of remote sensing image intelligent understanding,which is very important for practical application.As a carrier of maritime transportation,ships are an important tool for marine resources development,maritime transportation,marine military operations and other maritime activities.It is of the great significance to automatically detect them through optical remote sensing images.In this thesis,the ship detection method based on sketch information for optical remote sensing images is investigated.The main works are shown as follows:(1)A sea and land segmentation method for optical remote sensing images based on sketch information and superpixel segmentation is proposed.Firstly,according to the sketch map of the optical remote sensing image,the structural region map is generated,and the sea land classification is carried out on the structural region and the non-structural region in the structural region map according to the sea and land prior information,so the rough segmentation result map is obtained.Then,the superpixel segmentation is performed on the optical remote sensing image.Then the superpixels are classified by rough segmentation result map and the superpixels’ variance and local land-sea information.Finally,a superpixel two splitting algorithm is proposed to split the coastal land’s superpixels and the superpixels obtained by splitting are reclassified to realize the accurate positioning of the coastline and get the final result of land sea segmentation.Experimental results show that the proposed method has high segmentation accuracy and is robust to optical remote sensing images in different scenes.(2)An unsupervised ship detection method for optical remote sensing images is proposed,which is based on an improved Selective Search algorithm.There are the following problems when using Selective Search algorithm to extract ship candidate regions in optical remote sensing images: the number of candidate boxes is large,ships cannot be separated from side by side,a few candidate boxes can not completely surround ships.To solve these problems,the improved Selective Search algorithm is proposed to extract the candidate regions of ships in optical remote sensing images.The improved Selective Search algorithm can generate candidate regions for offshore ships and near-shore ships simultaneously with a high recall rate.Then,on the basis of those obtaining candidate boxes of the ship,in order to eliminate the false alarm and achieve the unsupervised detection of the ship,the candidate boxes are discriminated through voting method according to the prior information of the size,shape and sketch line segment distribution of the ship.Experimental results show that the proposed method can effectively reduce the number of ship candidate boxes and has good detection effect.(3)The precise positioning of the ships also is an important process of ship detection.Therefore,it is necessary to extract the ship’s contour based on the detection boxes.Here,a ship contour extraction algorithm based on sketch line segment analysis is proposed.Firstly,The algorithm uses the structural relationship between the sketch lines segment and the ship’s priori to filter and supplement sketch line segments,and extracts sketch lines segments in the periphery of ship.Then,rough contour of ship is obtained through the sketch line segment endpoint interconnection and the hole filling of image.Finally,the superpixel segmentation method is used to correct the contour of the ship to achieve accurate extraction of the ship’s contour.Experiments results show that the proposed algorithm is not affected by the internal objects of the ship,and can effectively extract high-quality ship contours.