Outlier detection in maritime environments using AIS data and deep recurrent architectures

TL;DR

This work tackles maritime anomaly detection using publicly available AIS data by employing an unsupervised encoder–decoder RNN framework to learn normal vessel motion and flag deviations via reconstruction error. The study compares Simple RNNs and GRUs, including bidirectional variants with recurrent dropout, finding that bidirectional GRUs offer superior temporal modeling and outlier discrimination. Using a 100-day AIS dataset from MarineCadastre NOAA data, the authors apply a six-sigma threshold on RMSE and a five-appearance rule to identify vessels with persistent or suspicious deviations, demonstrating practical utility for maritime surveillance. The approach avoids labeled data, is adaptable to different AIS sources, and lays groundwork for integrating additional data streams to enhance safety and security in maritime domains.

Abstract

A methodology based on deep recurrent models for maritime surveillance, over publicly available Automatic Identification System (AIS) data, is presented in this paper. The setup employs a deep Recurrent Neural Network (RNN)-based model, for encoding and reconstructing the observed ships' motion patterns. Our approach is based on a thresholding mechanism, over the calculated errors between observed and reconstructed motion patterns of maritime vessels. Specifically, a deep-learning framework, i.e. an encoder-decoder architecture, is trained using the observed motion patterns, enabling the models to learn and predict the expected trajectory, which will be compared to the effective ones. Our models, particularly the bidirectional GRU with recurrent dropouts, showcased superior performance in capturing the temporal dynamics of maritime data, illustrating the potential of deep learning to enhance maritime surveillance capabilities. Our work lays a solid foundation for future research in this domain, highlighting a path toward improved maritime safety through the innovative application of technology.

Figures (7)

• Figure 1: Schematic of Simple RNN, GRU models architecture without recurrent dropouts nor bidirectional layers.
• Figure 2: Schematic of SimpleRNN, GRU models architecture with recurrent dropouts in bidirectional layers.
• Figure 3: Root Mean Squared Error Histograms of Test dataset
• Figure 4: Outliers from Test data using the Bidirectional GRU model
• Figure 5: Ground Truth vs Prediction in a randomly selected vessel from Test data, for Latitude and Longtitude in a day's data using the Bidirectional GRU model