Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Learning and Trajectory Compression for Enhanced AIS Coverage

Thomas Gräupl, Andreas Reisenbauer, Marcel Hecko, Anil Rasouli, Anita Graser, Melitta Dragaschnig, Axel Weissenfeld, Gilles Dejaegere, Mahmoud Sakr

TL;DR

Coastal AIS coverage gaps limit maritime situational awareness. VesselEdge addresses this by combining M³fed federated learning for edge-based movement anomaly detection with BWC-DR-A anomaly-priority trajectory compression to operate over low-bandwidth links. The work presents a dual-edge architecture (far-edge vessels and near-edge coastal centers), demonstrates substantial data reduction while preserving critical anomalies on historical AIS data, and leverages the MobiSpaces data space for federated knowledge sharing. If scaled and validated in sea trials, VesselEdge could meaningfully enhance SaR operations and VTS effectiveness in bandwidth-constrained maritime environments.

Abstract

This paper presents the VesselEdge system, which leverages federated learning and bandwidth-constrained trajectory compression to enhance maritime situational awareness by extending AIS coverage. VesselEdge transforms vessels into mobile sensors, enabling real-time anomaly detection and efficient data transmission over low-bandwidth connections. The system integrates the M3fed model for federated learning and the BWC-DR-A algorithm for trajectory compression, prioritizing anomalous data. Preliminary results demonstrate the effectiveness of VesselEdge in improving AIS coverage and situational awareness using historical data.

Federated Learning and Trajectory Compression for Enhanced AIS Coverage

TL;DR

Coastal AIS coverage gaps limit maritime situational awareness. VesselEdge addresses this by combining M³fed federated learning for edge-based movement anomaly detection with BWC-DR-A anomaly-priority trajectory compression to operate over low-bandwidth links. The work presents a dual-edge architecture (far-edge vessels and near-edge coastal centers), demonstrates substantial data reduction while preserving critical anomalies on historical AIS data, and leverages the MobiSpaces data space for federated knowledge sharing. If scaled and validated in sea trials, VesselEdge could meaningfully enhance SaR operations and VTS effectiveness in bandwidth-constrained maritime environments.

Abstract

This paper presents the VesselEdge system, which leverages federated learning and bandwidth-constrained trajectory compression to enhance maritime situational awareness by extending AIS coverage. VesselEdge transforms vessels into mobile sensors, enabling real-time anomaly detection and efficient data transmission over low-bandwidth connections. The system integrates the M3fed model for federated learning and the BWC-DR-A algorithm for trajectory compression, prioritizing anomalous data. Preliminary results demonstrate the effectiveness of VesselEdge in improving AIS coverage and situational awareness using historical data.

Paper Structure

This paper contains 11 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. Experimental setup
  4. Method
  5. Federated Learning
  6. Trajectory compression
  7. Preliminary Results
  8. Anomaly Detection on the Far Edge
  9. Trajectory compression
  10. Anomaly Detection on the Near Edge
  11. Conclusion & Outlook

Figures (4)

  • Figure 1: VesselEdge aims to extend AIS coverage beyond coastal antennas by deploying equipped vessels at sea. The far-edge device (c and f) receives AIS messages and sends compressed, ML-prioritized trajectories to the near-edge infrastructure (a and e) at the coastal control room. Optionally, a second control room (d) can exchange ML updates with the first near-edge via the MobiSpaces data space (b).
  • Figure 2: AIS records (blue) and anomalies (red) detected by M³fed for tankers, cargo, and passenger vessels near Gothenburg harbor on 3rd July 2018.
  • Figure 3: AIS records (blue) and anomalies (red) post-compression with BWC-DR-A at bandwidth constraint that allows keeping $0.25$ of original points per window in average.
  • Figure 4: Average distortion in meters depending on the bandwidth constraint introduced by the BWC-DR (blue) and anomaly-aware BWC-DR-A (orange) compression. The bandwidth constraint is expressed on the x-axis as the average fraction of points kept in a window after compression.