AIS-CycleGen: A CycleGAN-Based Framework for High-Fidelity Synthetic AIS Data Generation and Augmentation

Automatic Identification System (AIS) data are critical for maritime domain awareness, enabling tasks such as vessel classification, trajectory prediction, and anomaly detection. However, AIS datasets frequently suffer from domain shifts, data sparsity, and class imbalance, which limit the generalization of predictive models. To address these challenges, this paper presents AISCycleGen, a novel data augmentation framework that leverages Cycle-Consistent Generative Adversarial Networks to synthesize realistic AIS sequences through unpaired domain translation. The proposed approach employs a specialized 1D convolutional generator with adaptive noise injection to enhance diversity while preserving spatiotemporal structure. AIS-CycleGen generates target-style synthetic data from source domains without requiring labeled target samples, effectively bridging distribution gaps. Comprehensive evaluations across multiple downstream tasks demonstrate that AIS-CycleGen improves domain alignment, enhances data diversity, and boosts the performance of learning models. These results position AIS-CycleGen as a robust and generalizable solution for augmenting AIS datasets in real-world maritime intelligence applications.