Enhanced YOLOv8 for accurate and efficient floating object detection on water surfaces.

Area of Science:

• Computer Vision
• Environmental Science
• Machine Learning

Background:

• Detecting floating objects on water surfaces is crucial for environmental monitoring and pollution control.
• Complex aquatic environments present challenges like reflections, noise, and dense debris, hindering accurate detection.

Purpose of the Study:

• To enhance object detection accuracy for floating debris in complex aquatic environments.
• To develop an efficient and robust model for real-time water surface monitoring.

Main Methods:

• Introduced SEDS-YOLOv8, a YOLOv8n variant incorporating Squeeze-and-Excitation (SE) attention, Distribution Shift Convolution (DSConv), and Enhanced Intersection over Union (EIoU) loss.
• Utilized a hybrid dataset of 28,000 images with data augmentation for model training and evaluation.
• Integrated SEDSConv for multi-scale feature extraction and SE attention to mitigate noise; EIoU loss improved localization.

Main Results:

• SEDS-YOLOv8 achieved 86.02% precision, 85.01% recall, and 88.82% mAP@0.5.
• The model demonstrates high efficiency with 2.90M parameters and 7.60 GFLOPs, outperforming the baseline YOLOv8n.
• Real-time inference was maintained at 103.7 FPS on NVIDIA RTX 4090 hardware.

Conclusions:

• Task-specific architectural adaptations in SEDS-YOLOv8 significantly improve water-surface detection accuracy.
• The model offers substantial accuracy gains without compromising computational efficiency.
• Publicly available code and datasets facilitate further research and application in environmental monitoring.