Realtime Monitoring of Animal Behavior Using Deep Learning Models
Accurate monitoring of animal health and behavior is crucial for improving welfare and productivity in livestock management. Traditional observation methods are time-consuming and prone to subjective bias. To address these challenges, we propose an automated system for behavioral pattern using deep...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Università degli Studi di Firenze
2025-07-01
|
| Series: | Journal of Agriculture and Environment for International Development |
| Subjects: | |
| Online Access: | https://www.jaeid.it/index.php/jaeid/article/view/16397 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate monitoring of animal health and behavior is crucial for improving welfare and productivity in livestock management. Traditional observation methods are time-consuming and prone to subjective bias. To address these challenges, we propose an automated system for behavioral pattern using deep learning-based pose estimation techniques. Specifically, we utilize ResNet-50, a deep convolutional neural network, to detect key anatomical landmarks such as the nose, eyes, ears, and body center. By tracking these keypoints, we generate movement trajectories that help identify behavioral patterns. For behavior classification, we initially applied a decision tree algorithm, achieving an accuracy of 60%. To enhance performance, we implemented a random forest classifier, which significantly improved the accuracy to 96%. The system tries to classify seven key behaviors: "stand," "sit," "eat," "drink," "aggressive," "sit with legs tied," and "let go of the tail." The random forest model achieved the highest accuracy in detecting "standing" and "aggressive" behaviors, while lower accuracy was observed for "eating" behavior. Additionally, our pose estimation model demonstrated high precision and recall metrics, indicating robust performance in keypoint detection with minimal deviation from ground truth annotations. This automated system reduces the need for manual observation and provides a reliable tool for monitoring some animal behavior. The potential applications extend to various domains, including animal studies and livestock management, offering a scalable and user-friendly solution for real-time behavior analysis.
|
|---|---|
| ISSN: | 2240-2802 |