YOLOv8n-DDSW: an efficient fish target detection network for dense underwater scenes

Area of Science:

• Aquaculture technology
• Computer vision
• Artificial intelligence

Background:

• Traditional aquaculture monitoring relies on manual sampling, which is labor-intensive and inefficient.
• Existing fish detection algorithms struggle in complex underwater environments due to occlusion and deformation.

Purpose of the Study:

• To develop an advanced fish target detection algorithm for complex intensive aquaculture scenarios.
• To address challenges like fish occlusion, deformation, and detail loss in underwater monitoring.

Main Methods:

• Proposed YOLOv8n-DDSW algorithm integrating C2f-deformable convolutional network (DCN) and dual-pooling squeeze-and-excitation (DPSE) attention.
• Incorporated small object detection and Wise-Intersection over Union (IOU) loss function.
• Trained and validated the algorithm on a Kaggle dataset.

Main Results:

• The improved YOLOv8n-DDSW algorithm showed significant gains: mAP50 increased by 3.9%, precision by 3.7%, recall by 6.1%, and mAP50-95 by 7.7% compared to the original YOLOv8n.
• Demonstrated enhanced detection accuracy for irregular and occluded fish targets.
• Improved performance in sensing small targets and bounding box regression.

Conclusions:

• The YOLOv8n-DDSW algorithm effectively solves low detection accuracy issues in intensive aquaculture.
• This technology supports the intelligent and modern development of fisheries through improved monitoring.
• The findings highlight the potential of AI in optimizing aquaculture operations.