Related Experiment Video
Updated: Jun 29, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
MFDSU-Net: a novel semantic-geometric collaborative network for cattle segmentation in complex farm environments.
Wangli Hao1, Yufan Jiang1, Yifei Liu1
1Faculty of Software Technologies, Shanxi Agricultural University, Jinzhong, Shanxi, China.
Frontiers in Veterinary Science
|June 26, 2026
Summary
Precision livestock farming
Area of Science:
- Computer Vision
- Agricultural Technology
- Machine Learning
Background:
- Precision livestock farming faces challenges in cattle segmentation due to background complexity and animal occlusion.
- Existing methods struggle with semantic uncertainty and geometric distortions, impacting accuracy.
Purpose of the Study:
- To introduce MFDSU-Net, a novel U-Net architecture for improved cattle segmentation.
- To enhance both semantic and geometric representations for robust segmentation.
Main Methods:
- Proposed MFDSU-Net architecture with Multi-Scale Feature Aggregation Block (MFABlock) and Multi-scale Deformation Sampling Block (MDSBlock).
- MFABlock enhances contextual understanding across scales.
- MDSBlock adaptively models spatial deformations to preserve boundaries.
Main Results:
- MFDSU-Net achieved a Dice score of 88.14% and an IoU of 81.38%.
- Outperformed state-of-the-art models by 0.5% (Dice) and 0.8% (IoU).
- Maintained high inference efficiency with 0.7M parameters and 2.8 GFLOPs.
Conclusions:
- MFDSU-Net effectively addresses cattle segmentation challenges in precision livestock farming.
- The model's efficiency makes it suitable for real-time deployment on edge devices.
