<p>With the rapid growth of maritime transportation, ensuring navigational safety has become increasingly critical. Accurate trajectory prediction is essential for anticipating vessel movements and reducing maritime risks. In recent years, most studies have focused on extracting spatiotemporal features from vessel trajectories. However, actual trajectory data often exhibit multi-scale features, making it difficult to fully capture their intrinsic patterns using spatiotemporal information alone. To overcome this limitation, we propose a trajectory prediction model based on time–frequency feature fusion. AIS data from waters near Denmark are used as the experimental dataset, with latitude, longitude, course over ground (COG), and speed over ground (SOG) used as both input and output features. A multi-scale frequency-domain analysis is employed to extract global trends and local variations, enabling effective integration of time–frequency features, reducing the risk of critical information loss, and maximizing the utilization of information. Finally, the model is evaluated against state-of-the-art methods on three real-world AIS datasets. The results demonstrate that it consistently achieves superior prediction accuracy across various scenarios.</p>