Research on marine flexible biological target detection based on improved YOLOv8 algorithm.

Area of Science:

• Marine Biology
• Computer Vision
• Image Processing

Background:

• Deformable marine biological entities present significant challenges for standard object detection algorithms due to foreground-background similarity.
• Underwater imaging often suffers from poor quality, further complicating the accurate identification of flexible subjects.

Purpose of the Study:

• To develop an advanced object detection algorithm specifically for identifying deformable marine flexible biological targets.
• To improve the accuracy and robustness of marine organism detection in underwater imagery.

Main Methods:

• A dataset of marine flexible biological subjects was compiled.
• Contrast Limited Adaptive Histogram Equalization (CLAHE) with boundary enhancement refined image quality.
• The YOLOv8n framework was modified with Deformable Convolutional Network (DCN) and RepBi-PAN modules.
• SimAM attention mechanism and Wise-IoU (WIoU) loss function were integrated.

Main Results:

• The enhanced algorithm demonstrated superior performance compared to the conventional YOLOv8n.
• Improvements were observed in capturing geometric transformations and processing essential features.
• The integration of CLAHE and WIoU addressed image quality and label issues, respectively.
• The modified network effectively handled foreground-background similarity and concentrated on pivotal areas.

Conclusions:

• The proposed algorithm significantly enhances the precision of marine flexible biological target detection.
• The combination of DCN, RepBi-PAN, SimAM, and WIoU offers a robust solution for underwater object detection.
• This work provides a valuable tool for marine research and conservation efforts.