Attention-enhanced and integrated deep learning approach for fishing vessel classification based on multiple features

Effective fisheries management is the key to achieve sustainable fisheries globally, while accurate monitoring of fishing vessels is essential to improve the effectiveness of management measures. Self-reported information on vessel types is often limited and may not cover all operating fishing vessels, causing incomplete monitoring in fisheries management. Therefore, a novel way to objectively identify the types of a large quantity of fishing vessels is needed. In this study, we presented an innovative integrated deep learning model by using automatic identification system (AIS) data to classify five types of fishing vessels, including gillnetter, hook and liner, trawler, fish carrier, and stow net vessel, further improving the performance of fishing vessel classification. First, we preprocessed data by removing erroneous information, dividing the vessel trajectories by day to obtain a complete and reliable dataset. Then, a multidimensional feature vector was constructed by combining the geometric, static and dynamic characteristics of fishing vessels to explain the behavioral differences of various types of fishing vessels more effectively. Finally, the feature vector was fed into an ensemble model of a two-dimensional bidirectional long short-term memory network and a convolutional neural network with an attention mechanism for training, and the prediction results were obtained through a fully connected layer. The accuracy of the ensemble model was 91.90%, which was higher than other single classifiers. The experimental results demonstrated that this method obtained remarkable performance and could be adopted to improve the precision of fishing vessel classification based on AIS data.