Deep-Sea Biological Detection Method Based on Lightweight YOLOv5n.

Area of Science:

• Marine Biology
• Computer Vision
• Deep-Sea Ecology

Background:

• Deep-sea biological detection is crucial for resource research and conservation.
• Poor image quality and limited data hinder current deep-sea detection models.
• Existing high-precision models are often too complex for deep-sea deployment.

Purpose of the Study:

• To develop an efficient and accurate deep-sea organism detection method.
• To address challenges of low contrast, color deviation, and insufficient data in deep-sea imagery.
• To create a lightweight model suitable for deployment in deep-sea environments.

Main Methods:

• A lightweight YOLOv5n architecture was developed.
• An image enhancement technique using global and local contrast fusion (GLCF) was integrated.
• A Ghost module and simAM-based Bottleneck (GS-Bottleneck) was designed for model efficiency.
• A transfer learning with knowledge distillation (TLKD) strategy was employed to reduce data dependency and improve generalization.

Main Results:

• The proposed method demonstrated significant improvements in detection accuracy and speed.
• The integrated GLCF enhanced image quality by addressing color deviation and low contrast.
• The GS-Bottleneck module ensured detection performance while maintaining model lightness.
• TLKD effectively improved model generalization and reduced data requirements.

Conclusions:

• The developed lightweight YOLOv5n with GLCF and TLKD offers a superior solution for deep-sea biological detection.
• The method overcomes limitations of existing models in terms of complexity and data dependency.
• This approach facilitates more effective deep-sea resource research and conservation efforts.