An Efficient Ship Detection Method Based on YOLO and Ship Wakes Using High-Resolution Optical Jilin1 Satellite...

Area of Science:

• Remote Sensing
• Computer Vision
• Maritime Surveillance

Background:

• Accurate ship detection in remote sensing is crucial for maritime security and monitoring.
• Conventional methods struggle with ships obscured by clouds or outside image boundaries.
• Existing techniques often have limitations in detecting partially visible or occluded vessels.

Purpose of the Study:

• To develop a practical and efficient ship detection scheme for remote sensing imagery.
• To enhance detection capabilities for ships, including those not fully visible.
• To provide a reliable method for identifying sailing direction and improving maritime situational awareness.

Main Methods:

• Combines deep learning for ship body detection with feature-based image processing for ship wake detection.
• Utilizes a deep convolutional neural network (CNN) for identifying ship hulls.
• Employs a feature-based approach to detect ship wakes, aiding in locating obscured vessels.

Main Results:

• Achieved over 93.5% success rate in detecting visible ships.
• Successfully detected over 70% of targets lacking a visible ship body by utilizing wake detection.
• Demonstrated high confidence and low false alarm rates in detecting ships using the proposed framework.

Conclusions:

• The proposed detection framework effectively addresses limitations of conventional methods for ship detection in remote sensing.
• The integrated approach of ship body and wake detection offers a robust solution for identifying partially obscured or boundary-exceeding vessels.
• This method significantly improves the detection of sailing ships in challenging remote sensing scenarios.