A CNN-LSTM Architecture for Marine Vessel Track Association Using Automatic Identification System (AIS) Data

Area of Science:

• Maritime surveillance
• Artificial intelligence
• Data science

Background:

• Distinguishing normal from anomalous vessel movement is crucial for maritime security.
• Traditional multi-object tracking algorithms struggle with spatial-temporal variations and data complexities in vessel tracking.
• Accurate vessel trajectory tracking is essential for timely threat detection and intervention.

Purpose of the Study:

• To develop an advanced framework for marine vessel track association.
• To address challenges posed by spatial-temporal variations, overlapping tracks, and missing data in vessel movement analysis.
• To improve the accuracy and efficiency of identifying potential threats in marine environments.

Main Methods:

• A novel 1D Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture was developed for track association.
• The framework treats vessel tracking as a multivariate time series problem, capturing spatial patterns and long-term temporal dependencies.
• The model was trained on Automatic Identification System (AIS) data, learning vessel trajectories from location and motion parameters.

Main Results:

• The proposed 1D CNN-LSTM framework demonstrated superior tracking performance compared to other neural network architectures.
• The model accurately associates sequential vessel observations, learning individual trajectories during training.
• Real-time vessel track output was achieved using AIS data, showing high accuracy, precision, recall, and F1 scores.

Conclusions:

• The 1D CNN-LSTM framework offers a robust solution for marine vessel track association, outperforming existing methods.
• This approach enhances the capability of marine surveillance systems to detect and monitor vessels effectively.
• The study highlights the potential of advanced neural networks in managing complex spatio-temporal data for maritime security applications.