Related Experiment Video
Updated: May 30, 2025
08:02
Author Spotlight: Innovative Ice Cream Melting Behavior Analysis Through a Computer Vision System
Published on: October 4, 2024
2.2K
A vision model for automated frozen tuna processing
Richeng Wang1, Xiongsheng Zheng2, Yan Chen3
1School of Marine Engineering Equipment, Zhejiang Ocean University, Zhoushan, 316002, People's Republic of China.
Scientific Reports
|January 25, 2025
Summary
TunaVision, a new YOLOv8-based model, accurately segments and estimates the pose of frozen tuna for automated processing. It overcomes challenges in size variation and complex multi-fish scenes, improving efficiency in seafood automation.
Area of Science:
- Computer Vision
- Robotics
- Agricultural Engineering
Background:
- Automated processing of frozen tuna requires accurate segmentation and pose estimation.
- Challenges include size variations, indistinct features, and complex multi-fish scenarios.
Purpose of the Study:
- To develop TunaVision, a vision model for automated frozen tuna processing.
- To enhance segmentation and pose estimation accuracy for frozen tuna.
Main Methods:
- TunaVision utilizes YOLOv8m-FusionSeg for improved instance segmentation.
- YOLOv8s RSF enhances feature extraction speed and accuracy for detection and localization.
- A vector-based approach is used for pose estimation.
Main Results:
- YOLOv8m-FusionSeg achieved an mAP@0.5 of 93.3% for segmentation.
- YOLOv8s RSF achieved an mAP@0.5 of 96.1% for detection.
- Pose estimation resulted in a Mean Absolute Error (MAE) of 1.81 degrees.
Conclusions:
- TunaVision effectively segments, detects, and estimates poses of frozen tuna.
- The model offers significant improvements over existing methods.
- Provides valuable insights for automated seafood processing systems.
