Application of Convolutional Neural Network (CNN) to Recognize Ship Structures

Area of Science:

• Maritime technology
• Artificial intelligence in logistics
• Robotics and automation

Background:

• Increasing use of drones for maritime logistics requires enhanced safety protocols.
• Accurate identification of vessels and structures is critical for autonomous drone navigation.
• Existing systems may lack the precision needed for complex shore-to-ship delivery environments.

Purpose of the Study:

• To develop and evaluate a system for recognizing ships and their structures to enhance drone delivery safety.
• To improve the reliability of autonomous drone operations in maritime settings.
• To support the first air delivery service by drones in Korea.

Main Methods:

• Utilized a convolutional neural network (CNN) for object detection and recognition.
• Employed the Detectron2 platform for CNN-based object sensing.
• Developed a dataset from the Marine Traffic Management Net for training and validation.

Main Results:

• The developed CNN system demonstrated effective recognition of ships and their structures.
• Performance metrics indicate high accuracy in distinguishing maritime objects.
• The system was validated through actual drone delivery operations.

Conclusions:

• The CNN-based recognition system significantly improves the safety of drone operations for shore-to-ship delivery.
• Accurate ship and structure identification is feasible with advanced AI models.
• This technology paves the way for safer and more efficient autonomous maritime drone services.