Related Experiment Videos
Underwater biological target detection algorithm based on DVS-YOLO11
Caiqin Jia1, Ling Wang2,3, Shengwen Gao2,3
1School of Computer Science and Technology, North University of China, Taiyuan, 030051, Shanxi, China. 20220076@nuc.edu.cn.
Scientific Reports
|July 20, 2026
Summary
This study introduces DVS-YOLO11, an enhanced YOLO11 algorithm for improved underwater target detection. The algorithm uses dynamic feature fusion and global context mechanisms, significantly boosting accuracy for challenging targets like scallops.
Area of Science:
- Computer Vision and Image Processing
- Robotics and Autonomous Systems
- Marine Technology
Background:
- Underwater environments present significant challenges for target detection due to light absorption, scattering, and resulting image degradation (blurring, color shifts).
- Existing target detection algorithms often struggle with accuracy and stability in these conditions, leading to false positives and missed detections.
Purpose of the Study:
- To enhance the accuracy and stability of underwater target detection systems.
- To develop an improved deep learning algorithm capable of overcoming the limitations of underwater imaging.
Main Methods:
- An improved YOLO11 algorithm, termed DVS-YOLO11, was developed incorporating several novel modules.
- A Dynamic Feature Fusion (DFF) mechanism was integrated into the C3k2 module (C3k2-DFF) to improve multi-scale feature representation.
- A Global Context (GC) mechanism was added to the Visual Feature Enhancement (VFE) module (VFE-GC) for adaptive feature recalibration and noise suppression.
- The Shape-IoU loss function was enhanced with Normalized Wasserstein Distance (NWD) (Shape-NWD) to improve bounding box regression for irregular targets.
Main Results:
- DVS-YOLO11 achieved a mean Average Precision (mAP@0.5) of 84.78% and mAP@0.5:0.95 of 64.67%, with 87.44% Precision and 75.57% Recall.
- Significant improvements were observed for scallop detection (mAP@0.5 increased from 63.9% to 68.6%), demonstrating effectiveness on targets with blurry edges and irregular shapes.
- The proposed method showed improvements across all metrics compared to the baseline YOLO11, while maintaining low computational costs.
Conclusions:
- The DVS-YOLO11 algorithm effectively addresses the challenges of underwater target detection by enhancing feature representation and localization accuracy.
- The integration of DFF, GC mechanisms, and the Shape-NWD loss function contributes to robust performance in complex underwater environments.
- The proposed method offers a promising solution for various underwater applications requiring reliable target detection.
