Development of Explainable AI-based Vessel Activity Monitoring for Korea Coast Guard Satellite Application

Summary

This project aims to provide explainable AI-based vessel activity monitoring service for Korea Coast Guard Satellite Application Project. The project includes runtime framework for maritime surveillance services that learn, deploy, and manage AI-based analysis models. It also performs service-specific analysis processing of real-time collected data and automates AI-based data management for new data that processes collected data by service.

The need for a more comprehensive surveillance system for the entire sea area has led to the development of a 24-hour monitoring system based on satellite imagery and automatic identification system (AIS) to overcome the limited coverage of existing maritime surveillance systems.  However, there are challenges with the satellite information service system and practicalization technology for practical use in the maritime police. To address these challenges, this project proposes a Vessel Activity Monitoring Subsystem in following three subjects:

  • Explainable AI-based Vessel Activity Monitoring Support: The XAI-based vessel activity detection support receives continuous input of EO/SAR imagery and AIS/V-PASS information from data streams and analyzes vessel activity based on Explainable AI (XAI), including identification of IUU and suspicious vessels, in the XAI runtime processing system.
  • Maritime Surveillance Service Runtime Processing Framework: We support the processing and serving of reference computation for maritime surveillance (ship activity and marine pollution surveillance) analysis models that receive satellite images and location information collected in real time. In addition, we developed a module to deploy and execute maritime surveillance analysis models and execution logic packaged in container-based units as microservices to the system.
  • AI data management automation framework: This approach aims to provide automated AI data refinement capabilities, including pre-processing of target vessel analysis for new input imagery (SAR, EO/IR) and measuring the quality of the data. Automate data management by using AI ship identification to detect and identify ships in new input imagery. Validate the suitability and harmfulness of the input data for the commercialization service model. Finally, request updates (retraining) of the commercialization service model due to changes in data versions.