Application Of Marine Target Detection Based On Deep Learning

With the rapid development of world economic globalization,maritime transportation has become more and more important.The increase in the number of ships has led to a gradual increase in accidents such as ship collisions and illegal smuggling.How to efficiently improve the transportation management of maritime traffic has gradually become a hot research issue.Effective management of ships first requires the efficient acquisition of the location and type of vessel targets at sea.Ship Automatic Identification System(AIS)and nearshore video surveillance system can provide rich basic data for ship management.AIS is widely used in ship position positioning and communication between ships,and has the characteristics of rich information and large amount of data.However,due to equipment failure and the interference of some human factors,there will be some abnormal data in the AIS data,and the existence of abnormal data will affect the effective management of ships.The data of nearshore video surveillance is mainly ship image data.In order to better manage the ship target,it is necessary to identify the type of the target ship in the image.This paper chooses to apply the deep learning algorithm to the task of marine target detection to process massive data with higher efficiency,thereby providing efficient solutions and ideas for solving the target detection of marine vessels.The main research contents of this paper are as follows:1.In order to extract the motion state information of ships contained in the AIS track data,a track clustering method based on deep learning is proposed.By establishing the LSTM-Autoencoder track clustering model,the clustering of AIS tracks is completed.Thus,the track clusters with similar sailing trends and sailing states in ships are obtained,which are used as the basis for subsequent track anomaly detection.The experimental results of the measured data verify the effectiveness of the LSTM-Autoencoder track clustering model.2.Based on the results of AIS track clustering,a method of track clustering based on deep learning is proposed.Based on the AIS track data in the track cluster,the algorithm establishes a CNN-LSTM track detection model for track prediction.On the basis of track prediction,the detection of abnormal track data is completed by setting a reasonable detection threshold.Finally,the proposed detection algorithm is verified by actual abnormal data.3.Aiming at the problem of image classification of near-shore video surveillance vessels,a deep learning CNN vessel type recognition model is built to complete the judgment and identification of vessel types.First,a corresponding data set is established based on the image data of near-shore video surveillance ships,and the data set is expanded by means of data enhancement.At the same time,a CNN ship type recognition model was established,and the Dropout regularization method was used to improve the robustness of the model,and the classification performance of the model was evaluated on the test data.Finally,a target ship type recognition system was designed and completed,which improved the recognition efficiency.