The Recognition Of The Ship Target In Sea Background Based On Gabor Feature Extraction

With the rapid development of national economy, a lot of money put into the development process of science and technology, computer technology and digital image analysis and processing technology have achieved the fly, In the information war, obtain information becomes particularly important because the accuracy of the information acquired directly affects the outcome of the war. As the importance of the navy in the modern military war rising and a variety of combat equipment with advanced combat ships constantly, and can be used to recognize the target and to become particularly important. Due to the complex sea background (the interference of the clouds and waves) and meteorological condition is not stable, lead to target imaging features are not obvious, which make ship target in infrared image recognition become a challenge. This thesis mainly studies the ship target recognition in complex background,complex background including the sky background,the surface of the background and some other interference background. Firstly analyzes the specific characteristics of the ship target, after understanding these basic knowledge, then further analyzes the factors that affect ship automatic target recognition, carefully studies the ship image preprocessing, ship target segmentation in the image, target feature extraction and combination, ship automatic identification system, and put forward the innovative point of view.Firstly, the thesis studied the background and significance of research related to this topic fundamentals ship target recognition, as well as the subject, to understand and analyze the current situation and abroad situation of the research,in order to put the focus on the study and solve the problem.Secondly, in view of the collected images are influenced by the gathering surroundings,need to get rid of the noise image preprocessing operations, fuse the improved Wavelet transform and Ridgelet transform experiments, and analysis the evaluation standards,according to the evaluation standard to judge the algorithm is effective or not. Aiming at low contrast,part of the target edge in the image blur,used the fuzzy contrast enhancement method to improve the contrast of the ship and the background, make the target contour pixels near the edge to up and down at the same time, and enhance target edges. Simulation results verify the effectiveness of the algorithm.Thirdly, for segmentation work, can be start with narrowing target area. This paper analyzes the background of air and water area is about equal to the sky, the gray value has certain difference, gray uniform distribution characteristics, so choosed PCNN model (pulse coupled neural networks model) combined with maximum shannon entropy criteria to implement the waters break up, on the basis of using the Hough transform to determine the antenna position, it can remove most interferential background that has nothing to do with the ship target ,then extract target characteristics to do the fusion, filter out noise, used the method of multiple feature fusion to highlight the ship targets in the images, then using the adaptive threshold to segment in the end. The segmentation results with the original image of the target relative ratios were observed for segmentation preserved degree target contour information, to verify the feasibility of the algorithm.At last, before the final identification of ship, ship required feature extraction, feature Moment commonly used to study and analyze the characteristics of less moment, and then the ship will be introduced Gabor feature feature extraction work, for extracting Gabor features high dimension is not conducive to follow-up calculations,using kernel principal component analysis method to compress dimension of the features, in the hope to make recognition faster and studied a kind of improvement of K-ELM method to study the target feature sequence that can be applied to a variety of different types of samples for identification, this method can have the better adaptability,smaller implementation error, and improve the stability.