Research on Fishing Vessel Recognition Based on Vessel Behavior Characteristics from AIS Data

Xiao Han ¹ Yang Zhou ^1,2,3* Jianjun Weng ^1,2 Lijia Chen ^1,2,3 Kang Liu ^1,2

• ¹ Wuhan University of Technology, Wuhan, China
• ² Hubei Key Laboratory of Inland Shipping Technology, Wuhan, China
• ³ State Key Laboratory of Maritime Technology and Safety, Wuhan, China

The Automatic Identification System (AIS) is one of the most important navigation assistance systems and plays a pivotal role in vessel monitoring. However, some fishing vessels disguise themselves as other vessel types during fishing bans to engage in illegal fishing activities, causing significant damage to marine ecosystem. To address this challenge and accurately identify vessel types, a BP-AdaBoost classification algorithm is developed by integrating backpropagation (BP) neural networks with ensemble learning techniques. The proposed algorithm leverages the AdaBoost method to combine multiple BP neural network weak classifiers into a strong classifier, effectively mitigating the slow convergence rate and susceptibility to local optima inherent in BP neural networks. By configuring the output nodes of the BP neural network to match the number of target classes, the AdaBoost algorithm achieves robust multi-class classification functionality. Historical AIS data are analyzed to extract static features, vessel behavior features, and temporal features for vessel classification. To minimize model overfitting, the Maximal Information Coefficient algorithm is employed to assess feature importance, and optimal feature combinations are determined through systematic feature selection experiments. Experiments are conducted using AIS data from the Pearl River Estuary in China, targeting the classification of cargo ships, fishing vessel, tanker, and passenger ships. The performance of the proposed method is compared with other machine learning algorithms. The results demonstrated classification accuracies of 90.8% for cargo ships, 95.6% for fishing vessels, 97.5% for tankers, and 98% for passenger ships, with an overall classification accuracy of 95%. Additionally, the BP-AdaBoost algorithm exhibited superior performance across other classification evaluation metrics. Specifically, the proposed algorithm outperformed the BP neural network by 4.5% and the support vector machine by 12.6% in overall classification accuracy. These findings indicate that the BP-AdaBoost algorithm is capable of effectively identifying vessel types based on historical trajectory data, providing a solid foundation for combating illegal fishing, detecting abnormal vessels, and identifying irregular vessel behaviors.