Related Experiment Video
Updated: Jun 13, 2026
07:34
Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies
Published on: November 7, 2025
Keypoint-Based Forest Musk Deer Behavioral Recognition Method
Dequan Guo1, Chuankang Chen1, Chengli Zheng2
1School of Automation, Chengdu University of Information Technology, Chengdu 610225, China.
Animals : an Open Access Journal From MDPI
|June 12, 2026
Summary
This study introduces an improved YOLOv8-Pose model for automated forest musk deer behavior recognition, enhancing conservation and breeding efficiency. The new method offers precise, real-time monitoring, overcoming limitations of traditional observation techniques.
Area of Science:
- Wildlife Biology
- Computer Vision
- Artificial Intelligence
Background:
- Traditional forest musk deer behavior monitoring is labor-intensive, subjective, and lacks real-time capabilities.
- Existing methods hinder efficient artificial breeding and effective wild population conservation efforts.
- There is a need for automated, accurate tools for monitoring endangered species behavior.
Purpose of the Study:
- To develop an advanced automated system for recognizing forest musk deer behaviors.
- To improve the efficiency and accuracy of monitoring for conservation and breeding programs.
- To provide a real-time behavioral analysis tool for endangered species protection.
Main Methods:
- Constructed a forest musk deer behavior image dataset with 18 keypoints annotated for four typical behaviors.
- Developed and integrated novel Dilated Spatial Pyramid Pooling-Fast (DILATED-SPPF) and Multi-scale Depthwise Separable Context Mixer (MDSC-Mixer) modules into YOLOv8-Pose.
- Evaluated the improved YOLOv8-Pose model against existing benchmarks for object detection and pose estimation.
Main Results:
- The improved YOLOv8-Pose model achieved superior performance in object detection (Box-mAP50: 0.929, Box-mAP50-95: 0.814) and pose estimation (Pose-mAP50: 0.879, Pose-mAP50-95: 0.565).
- The model significantly outperformed original YOLOv8-Pose and other comparison models.
- A visual interactive interface was developed for intuitive presentation of results.
Conclusions:
- The proposed method provides a high-precision, low-cost automated behavior analysis tool for forest musk deer.
- This technology significantly enhances the intelligence level of endangered species protection.
- The tool has substantial application value for both artificial breeding and wild conservation initiatives.
