Authors: Borja Carrillo Perez

Published on: October 07, 2024

Impact Score: 7.6

Arxiv code: Arxiv:2410.04946

Summary

• What is new: Introduces ShipSG dataset and ScatYOLOv8+CBAM architecture for real-time maritime ship recognition and georeferencing.
• Why this is important: Enhancing real-time maritime situational awareness through better ship recognition and georeferencing.
• What the research proposes: Custom deep learning model with advanced segmentation and attention mechanisms applied on embedded systems, combined with a novel slicing mechanism and georeferencing method.
• Results: Achieved 75.46% mAP, improved small and distant ship recognition by 8-11%, and positioning errors of 18-44 m.

Technical Details

Technological frameworks used: NVIDIA Jetson AGX Xavier, 2D scattering transform and CBAM

Models used: ScatYOLOv8+CBAM

Data used: ShipSG dataset with 3,505 images and 11,625 ship masks

Potential Impact

Maritime security firms, shipping companies, and port authorities could benefit from improved situational awareness and decision-making capabilities.

Want to implement this idea in a business?

We have generated a startup concept here: MarineVisionAI.