A Study On Harbor Target Recognition In High Resolution Optical Remote Sensing Image

At present, remote sensing information processing has been applied in many fields, and one important aspect is to monitor the harbor area, including the ships in the harbor. Harbor is the tie and also is the traffic hinge between land and sea. It is for parking ships, loading and unloading cargoes, and so on. As the development of the society, trade booms and passengers transport becomes frequent. In order to utilize the harbor effectively, monitoring the harbor area is very important and of significance. It involves the utilizing and supervising of the harbor, economy developing, nation security and many other aspects.The basic point of the paper is to monitor some important harbors through the optical remote sensing images. In this paper, a method of target detecting and recognition based on prior information of the harbors has been brought forward, after fully analyzing characters of the harbor target in the optical remote sensing image. Firstly, the method makes sure the accurate range of the harbor targets which need to be monitored in the input image of large size. Then, it detects and recognizes the ship targets which lie in the harbor area, according to the structure information of the harbor target.Recently, local features of image has become increasingly popular and been successfully applied in image processing, pattern recognition and other domains. In this paper, making use of the prior information of the harbor targets, SIFT feature matching was used to locate the harbor targets in the input images, and affine parameter, which was gained from the SIFT feature matching step, was used to match the prior structure information of the harbor from the prior harbor image onto the input remote sensing image accurately. It realized the harbor target recognition, avoiding image segmentation in traditional methods, and solved the question of pre-recognizing ship targets in the harbor area.In the part of ship targets detection, we fully utilized the docks distribution in harbors, and pick-up the directional information of the region where ships can berth itself in the harbor area. With the direct of the information, it is easily to get the ROI where ships mostly exist, and ships can be efficiently and truly detected. Especially, a method which is based on the static property of region along the contour of the ship was proposed in order to separate the ships which lay side by side. At last, a hierarchical control strategy was applied so that we can classify different type ship targets and get its particular information. In this step, type classification according to its actual use is done first, and then we employ image local feature to do in-depth recognition to get more information about the ship targets.