Researches For Ship Identification In Aerial Images

In information warfare, information has become the key to winning the war decision, processing all the information that received through various ways in a timely and rapidly manners, is the urgent problem. Target recognition is an important part of information technology, military surveillance and reconnaissance on the battlefield has an important role, because of its huge military application, the technology is research focus the international academic and the engineering community. Current the high-tech warfare that use information technology as the leading has a strong demand for automatic target recognition, detection and early warning, reconnaissance and surveillance, self-seeking enemy need automatic recognition technology for its support, thus laying target technology pivotal position in the high-tech military fields of world. Currently, the limitations of maritime reconnaissance and surveillance means to receive information at sea, mostly from satellite images, aerial images and radar images, Automatic identification of suspicious targets in the image instead of manual interpretation, is the development direction of imagery intelligence information processing technology.This paper firstly discusses the current theory, development and application situation of target recognition, then combining with the target image characteristics of ships, proposes and designs a ship automatic recognition system. Study in this paper include analysis of ship target image characteristics, image preprocessing, object segmentation algorithm and the improved, design target feature extraction, design recognition algorithm and implementation of system software and other aspects.Contrary to aerial images impacted by the acquisition environment, the contrast between target edge and background is poor, noise pollution and other situations, studies a variety of images preprocessing algorithms, propose contrast enhancement and filtering programs, experiments verify their effectiveness. According to the aerial images characteristics of ship targets, and based on a large number of experiments, design the Otsu threshold segmentation algorithm to complete target segmentation, then use neighborhood pixel information to improve its shortcomings, get target binary image that removing background information.Focus on the target feature extraction method, takes into account that the characteristics of ship shape was significant, analyze the effectiveness and feasibility of moment function describing shape feature, base on studying the advantages and disadvantages of the Hu invariant moments and affine invariants moment, propose a new combined moments as program of extracting feature, and test clustering ability of the amount of combined moments, prove the feasibility of this design.On the basis of depth analysis and comparison of the existing recognition algorithms, design the support vector machine classifier to perform classification. The method that compared with traditional methods has more capable generalization, eliminate the shortcomings of the curse of dimensionality, in the small sample and nonlinear case can reach a higher recognition rate. In this paper, the traditional SMO training algorithm has been improved effectively to reduce the number of iterations and improve the speed of training. Based on the previous study, develop target recognition applications software by VC the language, the software implements the proposed scheme, the results of test show that the software has good reliability and high recognition rate, the operation of the system is stable.