Table 1.
Abbreviations used in this paper.
Fig 1.
Input to the deep learning model and output trajectory.
Fig 2.
Process of real-time data gathered from the AIS device, as well as the trajectory extraction, AIS validity filtering, noise filtering, and the final preprocessed trajectory.
Table 2.
Statistical summary of AIS data.
Fig 3.
Preprocessed AIS trajectory data plotted as latitude versus longitude.
This visualization demonstrates the effectiveness of data preprocessing in producing a coherent representation of maritime movement patterns in the East Asian region.
Fig 4.
Conceptual structure diagram of the proposed model.
Fig 5.
Complete structure diagram of our proposed 1D-CNN model with historical data and target data.
Table 3.
Model performance for different ships.
Fig 6.
(a) to (n) shows the Actual trajectories and predicted trajectories for different ship IDs using CNN and DNN. (a) Ship ID: A0001—CNN, (b) Ship ID: A0001—DNN, (c) Ship ID: A0002—CNN, (d) Ship ID: A0002—DNN, (e) Ship ID: A0003—CNN, (f) Ship ID: A0003—DNN, (g) Ship ID: A0004—CNN, (h) Ship ID: A0004—DNN, (i) Ship ID: A0005—CNN, (j) Ship ID: A0005—DNN, (k) Ship ID: A0006—CNN, (l) Ship ID: A0006—DNN, (m) Ship ID: A0007—CNN, (n) Ship ID: A0007—DNN.
Fig 7.
Comparison of actual trajectories and predicted trajectories for different ship IDs using CNN and DNN.
Fig 8.
Perfromance comparison of each model with different ship ID.
Table 4.
Training and validation losses for ship IDs ABX-00004 and ABX-00005 using LSTM and GRU models.
Table 5.
Comparison of metrics across different methods.
Fig 9.
Perfromance comparison of each model.