Research on maritime ship target detection based on the optimized YOLOv8 model

Area of Science:

• Computer Vision
• Artificial Intelligence
• Marine Technology

Background:

• Accurate maritime ship detection is crucial for marine surveillance.
• Existing methods face challenges in diverse and complex marine environments.

Purpose of the Study:

• To develop an improved YOLOv8 model, named YOLOv8_optimize, for enhanced maritime ship detection.
• To increase the efficiency and robustness of ship detection systems.

Main Methods:

• Constructed a large dataset of over 80,000 annotated maritime images.
• Optimized the YOLOv8 backbone by integrating MBConv modules and depth-wise separable convolutions.
• Refined the detection head using Focal Loss to address class imbalance.

Main Results:

• The YOLOv8_optimize model demonstrated superior performance compared to YOLOv8n and YOLOv8s.
• Achieved significant reductions in computational complexity while maintaining high detection accuracy.
• Improved prioritization of difficult samples and rare ship categories.

Conclusions:

• YOLOv8_optimize offers an efficient and robust solution for maritime ship detection.
• The model has substantial practical value for marine surveillance applications.
• The applied optimizations enhance operational efficiency and detection performance.