YOLO-SEA: An Enhanced Detection Framework for Multi-Scale Maritime Targets in Complex Sea States and Adverse Weather

Area of Science:

• Computer Vision
• Artificial Intelligence
• Marine Technology

Background:

• Maritime object detection is crucial for safety and defense.
• Existing methods struggle with diverse targets and complex marine conditions.

Purpose of the Study:

• To develop an efficient maritime object detection framework.
• To enhance detection accuracy and robustness for various marine objects.

Main Methods:

• Utilized an enhanced YOLOv8 architecture named YOLO-SEA.
• Incorporated the SimAM-Enhanced SENetV2 Attention (SESA) module for feature representation.
• Implemented an improved Bidirectional Feature Pyramid Network (BiFPN) for multi-scale fusion.
• Employed Soft Non-Maximum Suppression (Soft-NMS) to reduce false detections.

Main Results:

• YOLO-SEA achieved a 5.8% improvement in mAP@0.5.
• Demonstrated a 7.2% improvement in mAP@0.5:0.95 compared to the baseline.
• Showcased enhanced accuracy and robustness in detecting eight maritime object types.

Conclusions:

• YOLO-SEA offers superior performance for maritime object detection.
• The proposed enhancements effectively address challenges in complex marine environments.
• This framework supports improved maritime surveillance and safety applications.