A transformer-based maritime foundation model trained on approximately 192 million AIS observations using self-supervised representation learning for vessel behaviour modelling, anomaly detection and operational intelligence. Designed to learn maritime behaviour directly from data, without labelled anomalies.
Maritime anomaly detection traditionally depends on manually defined behavioural rules or supervised learning requiring large labelled datasets. NAVISIGHT explores a different paradigm. Instead of learning anomalies directly, the model first learns what normal maritime behaviour looks like using self-supervised transformer representation learning. Behavioural deviations naturally emerge as anomalies within the learned latent representation space.
Million AIS Messages
Engineered Features
Transformer Dimension
Maximum AUROC
Transformer encoder trained using masked reconstruction.
Reconstruction + manifold drift anomaly detection.
Behavioural embedding retrieval.
Operational intelligence through latent analysis.
Representation AUROC
AIS Observations
Self-Supervised Learning
Journal Submission
Various pipelines and processes from data ingestion to inference
Screenshots showing overview of dashboard and various anomalies
NAVISIGHT A Self-Supervised Maritime Anomaly Detection Framework for AIS Trajectories Submitted to Ocean Engineering