Masterarbeit
Al-Driven Maritime Surveillance Using Multimodal Satellite Imagery for Vessel Detection
Completion
2026/1/
Research Area
Students
Chaitali Satishkumar Tamboliya
Advisers
Abubaker Gaber
Prof. Dr.-Ing. Martin Gaedke
Description
Maritime surveillance is essential for safeguarding economic zones, preventing illegal fishing, and ensuring national security. Traditional monitoring methods rely on manual analysis of satellite imagery and AIS (Automatic Identification System) data, which is time-consuming, error-prone, and unable to keep up with the growing scale of maritime activity. Additionally, a significant number of vessels, particularly those engaged in industrial fishing or illicit operations, operate without AIS, creating serious blind spots in current surveillance systems.
This master’s thesis investigates the development of an AI-powered system that integrates multimodal satellite imagery—Synthetic Aperture Radar (SAR) and optical images—with AIS data to automatically detect and verify vessels at sea. The approach involves training object detection models (e.g., YOLOv5, YOLOv8) separately on SAR and optical data. Their outputs are then fused and validated through cross-referencing with AIS signals to improve detection reliability and reduce false positives. Vessels detected without AIS matches will be flagged as potentially suspicious, considering known legal exceptions.Key challenges include synchronizing datasets with varying resolutions and timestamps, building a reliable labeled dataset using manual labeling through Roboflow and developing a method to assess the reliability of detections by checking if multiple models and AIS data agree. The system will be a web-based application to display detection results and collect expert feedback. The project will involve designing a data pipeline, training and evaluating deep learning models, and developing strategies for multimodal fusion and reliability assessment. Evaluation will be conducted using performance metrics such as precision, recall, and F1-score on a curated test dataset. This thesis aims to demonstrate how AI and multimodal data integration can enhance maritime monitoring by improving detection accuracy, reducing manual effort, and increasing trust in automated surveillance systems.
