Improved YOLOv8 Algorithm for Water Surface Object Detection.

Area of Science:

• Computer Vision
• Robotics
• Maritime Technology

Background:

• Surface target detection for uncrewed vessels faces challenges from scale variations and environmental factors like lighting and waves.
• Existing algorithms often struggle with accuracy, leading to false or missed detections in complex maritime settings.

Purpose of the Study:

• To enhance the accuracy and robustness of surface target detection for uncrewed vessels.
• To optimize the YOLOv8 algorithm for detecting water surface targets under challenging environmental conditions.

Main Methods:

• Proposed the YOLOv8-MSS algorithm, incorporating a small target detection head for improved sensitivity.
• Integrated C2f_MLCA in the backbone network to mitigate noise interference during downsampling.
• Utilized the SENetV2 lightweight model in the neck for better small target detection and anti-interference.
• Employed the SIoU loss function to improve bounding box regression precision and shape awareness.

Main Results:

• The YOLOv8-MSS algorithm achieved a mean Average Precision (mAP@0.5) of 87.9% on the FloW-Img dataset.
• Achieved an mAP@0.5:0.95 of 47.6%, demonstrating significant improvements over the original model.
• Reported performance gains of 5% in mAP@0.5 and 2.6% in mAP@0.5:0.95.

Conclusions:

• The YOLOv8-MSS algorithm effectively addresses scale differences and environmental noise in surface target detection.
• The enhancements lead to superior accuracy and reliability for uncrewed vessel maritime operations.
• The optimized model shows strong potential for real-world applications in maritime surveillance and autonomous navigation.