Research On Ship Detection And Identification Algorithm In High-Resolution Remote Sensing Images

With the rapid development of space satellite technology,the optical remote sensing image has achieved great progress in the acquisition and resolution.The application of ship detection technology in remote sensing images keeps drawing attention from military and civil.In this paper,we mainly focus on the ship detection and identification in optical remote sensing images under the background of sea and land.In order to improve the reliability of practice application,this paper mainly focuses on three parts: sea-land segmentation,ship detection and ship identification.We first present the coarse-to-fine process of sea-land segmentation.The coarse segmentation is accomplished by the OTSU combined with morphological method.The reallocation of isolated areas are fine segmented mainly based on the Euclidean distance.For the boundary between the land and sea that will bring a lot of false alarms,an effectively method is proposed to shield the land area.Firstly,the proposed method fills the land area with random value,which lies in 80% to 100% of peak value in histogram of the sea area.Secondly,a local sliding window filter is adopted to eliminate the boundary between the land and sea.In ship detection stage,a coarse-to-fine framework is proposed to detect ship candidates.During coarse detection process,Regions of interest(ROI)are extracted by combing the Top-Hat operator with visual saliency algorithm.In order to overcome the complex background of remote sensing images,a modified saliency fusion algorithm is derived to calculate saliency map.Then,fine detection is applied to deal with ROI based on shape feature and local information.Last but not least,an effective method based on the statistics change rule of wake along the ship driving direction is proposed to eliminate ship wake.In ship identification stage,an effective method based on the combination of local descriptors(SIFT,HOG,LBP)and Fisher Vector is proposed to represent optical remote sensing images with complex ship layout.The proposed algorithm reduces the influence of background by using local descriptors to represent local structure information of images.Then,the noise is reduced by using Fisher Vector to quantify the local features of ship.Lastly,linear SVM is applied to identify ship.The platform of ship detection and identification in high-resolution optical remote sensing image is accomplished by above algorithms,so as to realize the flexible man-machine interaction.Experiments on large number of images collected from Google Earth demonstrate the effectiveness of the proposed algorithms and reliability of the platform.