Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Sparsified Graph Learning Framework for Vessel Behavior Anomalies

Jeehong Kim, Minchan Kim, Jaeseong Ju, Youngseok Hwang, Wonhee Lee, Hyunwoo Park

TL;DR

The paper tackles vessel anomaly detection in dynamic maritime environments by introducing a sparsified graph learning framework that treats timestamps as explicit nodes to capture temporal dependencies. It builds a multi-ship graph within OPTICS-detected clusters, applies GCN-based encoding, and jointly optimizes forecasting and VGAE-based reconstruction to detect anomalies, with an edge-sparsification mechanism governed by an $L_0$-regularized mask. The approach contributes a novel timestamp-centered graph representation, an efficient sparsification strategy, and a dual forecasting-reconstruction objective, enabling robust detection even with limited labeled data. Practical impact arises from improved interpretability and scalability for real-time maritime surveillance using AIS data.

Abstract

Graph neural networks have emerged as a powerful tool for learning spatiotemporal interactions. However, conventional approaches often rely on predefined graphs, which may obscure the precise relationships being modeled. Additionally, existing methods typically define nodes based on fixed spatial locations, a strategy that is ill-suited for dynamic environments like maritime environments. Our method introduces an innovative graph representation where timestamps are modeled as distinct nodes, allowing temporal dependencies to be explicitly captured through graph edges. This setup is extended to construct a multi-ship graph that effectively captures spatial interactions while preserving graph sparsity. The graph is processed using Graph Convolutional Network layers to capture spatiotemporal patterns, with a forecasting layer for feature prediction and a Variational Graph Autoencoder for reconstruction, enabling robust anomaly detection.

Adaptive Sparsified Graph Learning Framework for Vessel Behavior Anomalies

TL;DR

The paper tackles vessel anomaly detection in dynamic maritime environments by introducing a sparsified graph learning framework that treats timestamps as explicit nodes to capture temporal dependencies. It builds a multi-ship graph within OPTICS-detected clusters, applies GCN-based encoding, and jointly optimizes forecasting and VGAE-based reconstruction to detect anomalies, with an edge-sparsification mechanism governed by an -regularized mask. The approach contributes a novel timestamp-centered graph representation, an efficient sparsification strategy, and a dual forecasting-reconstruction objective, enabling robust detection even with limited labeled data. Practical impact arises from improved interpretability and scalability for real-time maritime surveillance using AIS data.

Abstract

Graph neural networks have emerged as a powerful tool for learning spatiotemporal interactions. However, conventional approaches often rely on predefined graphs, which may obscure the precise relationships being modeled. Additionally, existing methods typically define nodes based on fixed spatial locations, a strategy that is ill-suited for dynamic environments like maritime environments. Our method introduces an innovative graph representation where timestamps are modeled as distinct nodes, allowing temporal dependencies to be explicitly captured through graph edges. This setup is extended to construct a multi-ship graph that effectively captures spatial interactions while preserving graph sparsity. The graph is processed using Graph Convolutional Network layers to capture spatiotemporal patterns, with a forecasting layer for feature prediction and a Variational Graph Autoencoder for reconstruction, enabling robust anomaly detection.

Paper Structure

This paper contains 23 sections, 19 equations, 2 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Spatiotemporal GNNs
  4. Vessel Behavior Anomaly
  5. GNN Based Vessel Anomaly Detection
  6. Methodology
  7. Problem Definition
  8. Graph Construction
  9. Graph Initialization
  10. Multi-ship Trajectory Graph
  11. Graph Sparsification
  12. Anomaly Detection
  13. Forecasting
  14. Reconstruction
  15. Anomaly Scoring
  16. ...and 8 more sections

Figures (2)

  • Figure 1: Our framework employs a multi-component architecture designed to learn a sparsified graph structure. The learned graph is then embedded using GCN layers, which supports both the forecasting and reconstruction processes. The combined loss from these components is utilized for effective anomaly detection.
  • Figure 2: Visualization of maritime vessel trajectories from the OMTAD dataset in Western Australian waters, one of the most congested maritime areas in the region. The movement paths show vessel positions over a 5-hour period, with lines representing trajectories and points indicating vessel positions. The OPTICS algorithm identified distinct vessel clusters (shown in different colors), while gray trajectories represent non-clustered vessels classified as noise.