Context-Aware Autoencoders for Anomaly Detection in Maritime Surveillance

TL;DR

This work tackles context-dependent anomalies in maritime AIS data and introduces a context-aware autoencoder framework to model context-specific normality. It analyzes four architectures—AE, MoE-AE, CAE, and GCAE—with context-specific thresholds, demonstrating that incorporating vessel type and navigational status substantially improves detection of collective and contextual anomalies. The results show that CAE and MoE-AE outperform a conventional AE in accuracy, while GCAE provides substantial model-size reductions with minimal loss of detection performance (e.g., ~20% fewer parameters with 86% overlap against CAE). The proposed approach offers a scalable, practical solution for real-world maritime surveillance, enabling more robust anomaly detection under variable sensor coverage and context shifts. The study also defines precise thresholding strategies, such as $\tau_c=\mu_c+\lambda\sigma_c$ with $\lambda=5$, and demonstrates the importance of context-aware modeling for reliable anomaly detection in AIS trajectories.

Abstract

The detection of anomalies is crucial to ensuring the safety and security of maritime vessel traffic surveillance. Although autoencoders are popular for anomaly detection, their effectiveness in identifying collective and contextual anomalies is limited, especially in the maritime domain, where anomalies depend on vessel-specific contexts derived from self-reported AIS messages. To address these limitations, we propose a novel solution: the context-aware autoencoder. By integrating context-specific thresholds, our method improves detection accuracy and reduces computational cost. We compare four context-aware autoencoder variants and a conventional autoencoder using a case study focused on fishing status anomalies in maritime surveillance. Results demonstrate the significant impact of context on reconstruction loss and anomaly detection. The context-aware autoencoder outperforms others in detecting anomalies in time series data. By incorporating context-specific thresholds and recognizing the importance of context in anomaly detection, our approach offers a promising solution to improve accuracy in maritime vessel traffic surveillance systems.

Figures (4)

• Figure 1: AE, MoE-AE, CAE & GCAE architectures (*CAE has one decoder per context / GCAE has one per group of contexts)
• Figure 2: Contexts in the dataset, each with associated vessel type.
• Figure 3: Validation reconstruction loss distributions per decoder and context. Each subplot corresponds to one decoder branch, with samples color-coded by context. The red dashed line represents the context-specific anomaly detection threshold $\tau_c$ computed using the $5\sigma$ rule. This highlights the intra-context variability and supports context-aware thresholding..
• Figure 4: Test dataset reconstruction losses obtains with CAE on multiple decoders for context $c_{21}$ data (drifting longlines vessels under way sailing) through multiple decoders.