Evaluation of AI Tools in Maritime Environments (Objectives #1 & #2)
SEAGUARD evaluated and selected AI tools for maritime object detection, classification, and anomaly detection. Since most AI vision research targets terrestrial environments, WP5 focused on adapting and benchmarking methods for maritime use. Among more than 10 tested algorithms, Vision Transformers (ViT) showed the best object detection performance, while YOLO proved effective for classification and localization.
Acoustic vessel classification remains limited by small, heterogeneous datasets. To address this, a method was developed to calibrate and merge diverse hydrophone datasets into a unified corpus, including data augmentation with varied ambient noise. A cepstrum-based ranging technique was also created to estimate target distance from single hydrophones, enabling better fusion with AIS, radar, and camera systems.
Additional work included deploying YOLOv26 (January 2026), applying deep neural networks and Isolation Forest for anomaly detection, and performing detailed AIS trajectory analysis (e.g. Haversine-based calculations). A Composite Maritime Anomaly Detection Quality Index was introduced for AIS model evaluation. A flexible Streamlit-based dashboard was also developed to support stakeholders and enhance operational performance.
SEAGUARD Interoperability Framework (Objective #4)
WP7 finalized the SEAGUARD Interoperability Framework (S.IF) defining rules, protocols, and data formats that enable heterogeneous entities to exchange information and collaborate within a shared digital federation.
S.IF establishes a virtual interoperability space where services, systems, and applications from different authorities can integrate, share data, and build joint capabilities. Entities joining the federation must comply with S.IF standards, becoming S.IF nodes via connectors or gateways, allowing them both to contribute and access services and data.
The framework includes two main functionality sets: Core functionalities for managing the interoperability space, and AI functionalities for mission awareness and decision support. The resulting architecture is flexible, secure, and adaptable to diverse maritime security scenarios.
Docking (Objective #3)
Docking/landing of unmanned systems (WP9) is an active research topic, especially in maritime environments. Contributions to the scientific literature and to the technological landscape are expected. During this reporting period, a review of the state-of-the-art was performed to set the foundations for future developments. A systematic approach was presented, identifying all relevant building blocks that are common to UxVs, regardless of their domains. A preliminary architecture was developed and the main components were identified. These developments pave the way for potentially innovative solutions. A novel approach to UUV docking is under development.
