Research On Ship Target Event Detection And Intelligence Analysis Technology

Ship is the key target for maritime monitoring,management and combat.Fast and accurate detection of typical events on ship targets,combined with intelligence analysis to identify behavior of ship targets,can provide support for various disposals and decisions such as control of maritime traffic,protection of marine exclusive economic zone,regulation of illegal fishing,and crackdown on smuggling.At present,there are some shortcomings in the process of ship target detection and intelligence analysis such as low degree of intelligence,inability to process information in real time,inefficiency of application mode,and so on.How to use marine big data to achieve ship target event detection and intelligence analysis has become the research focus and difficulty of development of marine information intelligent processing.So,aiming at the requirements of maritime big data intelligent processing,research on key technologies related to ship target event detection and intelligence analysis based on deep learning is carried out in this paper.The main works are as follows:1)Through the analysis of current research situations,AIS data preprocessing methods and deep learning-related theory,a ship target event detection method based on deep learning is designed and implemented in this paper.The existing convolutional neural network is used as the basic network,and the method proposed improves and optimizes by learning branch of sequential information,channel weighting of trajectory feature,multi-scale convolution layer and fast connection,which enhances the feature expression ability of deep network,the capture ability of important features and the adaptability of local variation of features.2)Based on the background of evidence collection and supervision of illegal fishing in offshore marine pastures,ship target event detection and intelligence analysis method is further designed and improved.Firstly,the optimal estimation of ship target can be obtained by maritime multi-source heterogeneous big data information fusion.Then,combined with the deep learning-based ship target event detection method based and the slide window-based variable length trajectory analysis algorithm,the problem that the event trajectory is difficult to accurately divide is solved.Finally,the event analysis method based on event detection is used to achieve the identification of illegal fishing events.3)Experimental analysis and discussion are carried out.Firstly,using the published real AIS data,the ship target feature trajectory data set is constructed.Secondly,under the same identification framework of ship target event detection,the performance comparison experiments are completed.Experimental results prove that the method proposed in this paper achieves 90.8% of accuracy,which is better than the other seven existing methods,and the method achieves 5.6 percentages point improvement compared with the basic convolutional neural network.Thirdly,the results of different events show that the recall rate of 95.1% for fishing events is obtained by the method in this paper,which meets the needs of the application of forensic and regulatory in illegal fishing.Finally,the method is deployed in the specified embedded device,and the processing speed of 200 segments per second is realized,which meets the real-time requirement.4)Based on the front-end separation framework of Vue.js and Spring Boot,using Java,Mybatis and other tools,the ship event detection prototype system for maritime intelligent supervision application is designed and implemented.