Related Experiment Video
Updated: Jun 28, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Deep learning framework for bovine iris segmentation.
Heemoon Yoon1, Mira Park1, Hayoung Lee2
1School of Information Communication and Technology, University of Tasmania, Hobart 7005, Australia.
Journal of Animal Science and Technology
|April 15, 2024
Summary
This study introduces a deep learning framework for bovine iris segmentation, optimizing animal biometrics for livestock traceability. The U-Net with VGG16 backbone achieved 99.50% accuracy, even with minimal annotated data.
Area of Science:
- Computer Vision
- Machine Learning
- Animal Biometrics
Background:
- Iris segmentation is crucial for livestock traceability systems.
- Accurate animal identification relies on robust biometric data.
Purpose of the Study:
- To develop a deep learning framework for pixel-wise bovine iris segmentation.
- To optimize segmentation performance using minimal annotation labels.
- To establish a reliable deep neural network (DNN) training framework.
Main Methods:
- Utilized the BovineAAEyes80 public dataset.
- Trained 15 deep neural network (DNN) models with varied encoder backbones and segmentation decoders.
- Evaluated models using accuracy, dice coefficient, and graphical results.
Main Results:
- Identified U-Net with a VGG16 backbone as the optimal model.
- Achieved 99.50% accuracy and 98.35% dice coefficient score.
- Demonstrated accurate segmentation of corrupted images with limited annotations.
Conclusions:
- The proposed framework effectively segments bovine irises with high accuracy.
- The U-Net with VGG16 backbone is optimal for this task.
- This advancement supports reliable DNN training for livestock biometrics.
