BGLE-YOLO: A Lightweight Model for Underwater Bio-Detection

Area of Science:

• Computer Vision
• Marine Biology
• Robotics

Background:

• Underwater environments present significant challenges for object detection due to low contrast and chromatic aberration.
• Automated methods are crucial for accurate identification of small objects in underwater imagery.

Purpose of the Study:

• To develop an efficient and accurate underwater object detection model suitable for edge devices.
• To address the limitations of existing models in detecting small and low-contrast underwater objects.

Main Methods:

• Introduction of an efficient multi-scale convolutional EMC module to enhance the backbone network.
• Integration of a global and local feature fusion module for small targets (BIG) into the neck network.
• Construction of a lightweight shared head (LSH) with reparameterization to maintain accuracy.

Main Results:

• The BGLE-YOLO model demonstrates high accuracy on the DUO and RUOD underwater datasets.
• Achieved comparable accuracy to benchmark models with significantly reduced computational cost (6.2 GFLOPs).
• The model boasts an ultra-low parameter count of 1.6 MB, ideal for edge deployment.

Conclusions:

• BGLE-YOLO provides an effective solution for automated underwater object detection, particularly for small and challenging targets.
• The model's lightweight design and high efficiency make it suitable for real-time applications on edge devices.
• This research contributes to advancing computer vision applications in marine research and underwater robotics.