Related Experiment Video
Updated: Sep 13, 2025
06:19
Integration of Animal Behavioral Assessment and Convolutional Neural Network to Study Wasabi-Alcohol Taste-Smell Interaction
Published on: August 16, 2024
535
Automated Cattle Head and Ear Pose Estimation Using Deep Learning for Animal Welfare Research
1Laboratory of Large Animal Clinical Medicine, Graduate School of Veterinary Sciences, Osaka Metropolitan University, Osaka 598-8531, Japan.
Veterinary Sciences
|July 25, 2025
Summary
This study introduces an AI system using deep learning to objectively assess cattle welfare by analyzing head and ear posture. The technology provides accurate, quantitative data for improved animal stress monitoring and farm management.
Area of Science:
- Animal Behavior and Welfare Science
- Computer Vision and Artificial Intelligence
- Biotechnology and Agricultural Engineering
Background:
- Animal welfare is increasingly important, with behavioral indicators like head and ear posture used to assess emotional state and stress.
- Traditional visual observation methods for animal behavior are subjective and unsuitable for long-term, quantitative monitoring.
- Objective, non-invasive methods are needed for reliable animal welfare assessment in field conditions.
Purpose of the Study:
- To develop and validate an artificial intelligence (AI)-based system for the detection and 3D pose estimation of cattle heads and ears.
- To enable objective, quantitative, and long-term monitoring of cattle behavior and stress levels.
- To provide a practical tool for animal welfare research and real-time farm management.
Main Methods:
- Utilized deep learning techniques, integrating Mask R-CNN for object detection and FSA-Net for 3D pose estimation (yaw, pitch, roll).
- Developed comprehensive datasets using images of Japanese Black cattle under natural conditions, annotated for detection and pose estimation.
- Evaluated system performance using metrics such as mean average precision (mAP) and mean absolute error (MAE).
Main Results:
- Achieved high performance with mAP of 0.79 for head detection and 0.71 for left ear detection.
- Demonstrated robust 3D pose estimation with a mean absolute error of approximately 8-9° across diverse cattle orientations.
- The AI system showed reliable performance in detecting and estimating the pose of cattle heads and ears.
Conclusions:
- The AI-based system offers a significant advancement over subjective visual observation methods for assessing animal welfare.
- Enables objective, quantitative, and long-term monitoring of cattle behavior, crucial for welfare assessment.
- The developed system has strong potential for practical applications in animal welfare research and precision agriculture.
