The Ship Detection Based On Optical Remote Sensing Images

As human beings pay more attention to the sea, ship becomes the crucial tool to utilize and explore sea. In order to get more information of ship, the ship detection technology was developed and applied to the area about national security, such as monitor traffic, fisheries, succor of the sea , anti-terrorism and anti- pirate.Compared the traditional ship detection based on SAR, the ship detection based on the optical images was more difficult because of the imaging of the optical sensors, while the related papers ignored.Though the research of papers and documents, we main discussed three problems:1. Optical images had shadings because of the rays changed, we can't use one threshold to segment the sea and the land.2.Optical images had many clouds, the similarities between some clouds and the sea&land area were hard to identify. So, the cloud detection algorithm was hard to design.3.We can't distinguish the ship target from the complex sea surface background, especially for some dark ship target.The problems above would impact the false alarm rate and missed alarm rate. Aiming at that problems, we designed the ship detection algorithm based on optical images in terms of study the related papers. The algorithm was finally used in the GoogleEarth platform, had the higher detection rate compared with the traditional algorithm.Our main work and contribution as follow shows:1. We proposed a new segmentation algorithm of sea and land based on 3D reconstruction aiming at the change of shading. Using the shape of shading method to get height information, than combining shape filters and the threshold segmentation to segment sea and land. Compared with the traditional 2D algorithm, our algorithm portrayed the land area better and was more robust.2. Aiming at the cloud detection, we used the image learning of content based method, extracted eigenvector from luminance component, texture, and edge information of sub-region in spatial domain, and used the SVM to train classification model and classify cloud. It also had feedback mechanism to adjust the parameters of the model, which improved the accuracy and the robustness of cloud detection.3. We finally proposed a new ship detection approach for optical remote sensing images via multi-level vision perception and vision attention allocation mechanism. First this algorithm extracted the spectral residual of an image and got full resolution saliency maps; then extracted the interesting targets, computed the interesting points; Using Gabor filters of adaptive orientation to get more information and describing the targets. Experiment results showed that our method can more effectively detect ship targets compared the traditional methods, especially for the complex sea surface background.In the end of this thesis, we concluded our work and discussed the plan in the future.