AIS Data Driven CNN-BiGRU Model for Ship Target Classification

Abstract

Most traditional methods for classifying marine ship targets using trajectory data rely on manual feature extraction, and it is difficult to consider the influence of spatial and temporal features on the classification results. In this paper, we propose a combination of a convolutional neural network and bidirectional gate recurrent unit (CNN-BiGRU) to classify ship targets using automatic identification system (AIS) data. First, AIS data are pre-processed to obtain valid ship trajectory segments, and basic information of the trajectory points, such as the speed, heading, and time, are used to construct the input feature vectors for the CNN and BiGRU, respectively. Second, the best CNN is trained and combined with BiGRU to obtain the CNN-BiGRU model. The combined model then simultaneously mines the spatio-temporal features contained in the AIS data and fuses the learned deep-level features. Finally, a fully connected layer is used to obtain the classification results of the ship targets. Compared with traditional machine learning algorithms and single deep learning models, the classification method in this study only requires the construction of simple input feature vectors and has different degrees of improvement in the four evaluation indexes of classification accuracy, precision, recall, and f-score, indicating that this method can effectively combine spatio-temporal features to improve the classification effect.