BSE-YOLO: An Enhanced Lightweight Multi-Scale Underwater Object Detection Model.

Area of Science:

• Computer Vision
• Robotics
• Marine Technology

Background:

• Underwater images present significant challenges including low contrast, blurred targets, and object clustering.
• Traditional object detection methods struggle with accuracy and are computationally intensive for underwater robots.

Purpose of the Study:

• To develop a lightweight and accurate object detection model for underwater applications.
• To address the limitations of existing methods in constrained robotic environments.

Main Methods:

• Proposed an enhanced lightweight YOLOv10n model named BSE-YOLO.
• Replaced the original neck with an improved Bidirectional Feature Pyramid Network (Bi-FPN) to reduce parameters.
• Introduced a Multi-Scale Attention Synergy Module (MASM) and integrated Efficient Multi-Scale Attention (EMA) for enhanced feature perception and fusion.

Main Results:

• BSE-YOLO achieved 83.7% mAP@0.5 on URPC2020 and 83.9% mAP@0.5 on DUO.
• The model reduced parameters by 2.47 M while improving mAP@0.5 by 2.2% and 3.0% over YOLOv10n.
• Demonstrated a significant reduction in parameters (approx. 0.2 M) compared to the baseline YOLOv10n.

Conclusions:

• BSE-YOLO offers an effective balance between accuracy and lightweight design for underwater object detection.
• The proposed model provides a viable solution for object detection tasks on underwater robots with limited resources.