Related Experiment Video
Updated: Aug 8, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
596
In Situ Sea Cucumber Detection across Multiple Underwater Scenes Based on Convolutional Neural Networks and Image
Yi Wang1, Boya Fu2, Longwen Fu2
1Coastal Defense College, Naval Aeronautical University, Yantai 264003, China.
Sensors (Basel, Switzerland)
|February 28, 2023
Summary
This study enhances underwater sea cucumber detection using artificial intelligence. Non-local image dehazing significantly improved the accuracy of YOLOv7 and YOLOv5 models for marine surveys.
Area of Science:
- Marine Biology
- Computer Vision
- Artificial Intelligence
Background:
- Traditional marine surveys are labor-intensive and being replaced by AI.
- Underwater image distortion and degradation hinder accurate object detection.
Purpose of the Study:
- To develop an automatic sea cucumber monitoring system using YOLOv7.
- To improve detection accuracy by addressing underwater image quality issues.
Main Methods:
- Implemented YOLOv7 for sea cucumber detection.
- Applied five image enhancement techniques, including Non-local image dehazing (NLD).
- Evaluated detection performance of YOLOv7 and YOLOv5 with and without enhancement.
Main Results:
- Non-local image dehazing (NLD) proved most effective for sea cucumber detection.
- YOLOv7 with NLD achieved the highest average precision (AP) of 0.940.
- NLD enhanced YOLOv7 and YOLOv5 AP by 1.1% and 1.6%, respectively.
Conclusions:
- Image enhancement, particularly NLD, significantly boosts AI-based sea cucumber detection accuracy.
- YOLOv7 demonstrates real-time performance (4.3 ms prediction time) suitable for marine organism surveying.
- The proposed method is applicable to underwater mobile platforms and video analysis.
