Sea Mine Detection Framework Using YOLO, SSD and EfficientDet Deep Learning Models.

Area of Science:

• Maritime safety
• Artificial Intelligence
• Computer Vision

Background:

• Geopolitical conflicts have increased the threat of sea mines in navigation routes.
• Drifting sea mines pose a significant risk to commercial and tourism activities in Black Sea countries.
• Deep learning is increasingly utilized in military operations, including threat detection.

Purpose of the Study:

• To develop and evaluate a deep learning system for detecting floating and underwater sea mines.
• To address the challenge of limited sea mine imagery through data augmentation and synthetic generation.
• To assess the performance of different deep learning models for sea mine detection.

Main Methods:

• Utilized image data from various sources (drones, submarines, ships).
• Employed data augmentation and synthetic image generation to create datasets for floating and underwater mines.
• Trained and compared three deep learning models: YOLOv5, SSD, and EfficientDet.

Main Results:

• The developed system achieved high accuracy in recognizing and detecting both floating and anchored sea mines.
• YOLOv5 and SSD models were effective for floating mine detection, while YOLOv5 and EfficientDet excelled in underwater mine detection.
• Tests on portable equipment like Raspberry Pi indicated feasibility for real-time applications.

Conclusions:

• Deep learning models, particularly YOLOv5, SSD, and EfficientDet, show significant promise for accurate sea mine detection.
• Data augmentation and synthetic data generation are effective strategies for overcoming limited datasets in this domain.
• The system's potential for real-time deployment on portable computing devices enhances maritime security capabilities.