Research On Marine Ship Target Detection Method

In recent years,with the rapid development of science and technology,the global marine strategy has become more and more intense.The ocean has become an indispensable place for military affairs of all countries and an important channel for global trade and commerce.At present,due to the frequent occurrence of military activities at sea and the rising volume of maritime transportation,the maritime navigation environment is becoming more and more complex,and maritime traffic safety accidents occur frequently.With the development of sensor technology and video technology,it is of great practical significance and application value to real-time monitor the sea targets by video means.At present,the development of marine ship video surveillance system is still in its infancy,mainly due to the complex sea environment and many uncertainties of ship targets.At present,the difficulty is to detect ship targets in real-time and accurately in complex and changeable marine environment,and lays an important foundation for subsequent recognition,classification and tracking of ship targets.In this paper,a deep research on the ship target detection algorithm in marine video surveillance,and three target detection algorithms working on complex sea conditions and ship targets’ characteristics ware proposed.The main research works were as follows:In order to prove the algorithms of ship target detection in this paper,a ship targets data set was established,and the data set was classified according to the experimental requirements.For the influence of maritime noises,an improved Canny edge detection operator was proposed.This algorithm could suppress maritime noises and retain more edge information very well by using compound morphological filter.It used Otsu method to set the threshold adaptively,and used mathematical morphological algorithm to post-process the target edge.Experiments proved that the algorithm improved the precision of the ship targets’ edge.For the size and angle change of ship target during the voyage,a multi-scale Harris-Laplace corner detection algorithm which based on edge feature was proposed.Based on the edge of the ship targets ware extracted by the improved Canny algorithm,the corners of the ship targets were extracted by the improved multi-scale Harris-Laplace corner detection algorithm.The convex hull of the ship targets were extracted by the improved Graham algorithm.Experiments proved that the algorithm improved the precision and real-time of the automatic extraction corners and detection the ship targets.For the maritime noises and cloud-mist weather,an algorithm named ViBeDiff5 for detecting ship targets was proposed.The algorithm based on five frame difference algorithm and improved ViBe algorithm.The moving region of the ship targets were obtained by the improved five-frame difference method,and improved the ViBe algorithm in the aspects of background model initialization and updating.ViBeDiff5 combined these two methods could get more accurate and complete information of the ship targets,at the same time could eliminated "ghost" quickly and greatly suppressed various kinds of noise.Thus,the accuracy and robustness of ship target detection algorithm could be improved.Finally,the ship targets were extracted from the binary image of ship target detection results by the minimal outer rectangle.Experiments proved that the algorithm improved the location precision of the ship target.For the small targets in fog,an improved method of ship target detection based on visual attention mechanism was proposed.The algorithm used Otsu method and Hough transform to extract the sea-horizon,and delimit the sea-horizon region as the extraction range of the ship targets.Based on the classic Itti model,the improved wavelet transform was used to extract the high-frequency and low-frequency features.The improved Gabor filter and DMT were used to obtain the direction and edge texture features respectively.The color and motion features were extracted in HSI color space,Then the saliency maps were obtained by weighted linear fusion,and then the ship target regions were segmented.Experiments proved that the algorithm had high detection precision and low redundancy,and showed strong adaptability in the case of small targets in fog.