Related Experiment Video
Updated: Apr 30, 2026
05:41
A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
Published on: February 6, 2020
8.7K
BCST-GCN: a skeleton-based spatiotemporal graph convolutional network with bidirectional cross-attention for pig
Haojie Chai1, Weibo Zhan2, Jianshuai Su1
1College of Artificial Intelligence, Henan Institute of Science and Technology, Xinxiang, China.
Frontiers in Veterinary Science
|April 29, 2026
Summary
This study introduces a new skeleton-based method for pig behavior recognition, improving accuracy and robustness. The advanced model enhances spatiotemporal feature extraction for intelligent farming applications.
Area of Science:
- Computer Vision
- Animal Behavior Analysis
- Machine Learning
Background:
- Existing video-based pig behavior recognition methods struggle with weak inter-frame motion correlation and poor robustness.
- Current approaches often fail to fully exploit skeleton spatiotemporal dynamic features and capture fine behavioral details.
Purpose of the Study:
- To propose a novel skeleton-based spatiotemporal dynamic modeling method for enhanced pig behavior recognition.
- To improve the accuracy, precision, and recall in recognizing pig behaviors like feeding, walking, and lying.
Main Methods:
- Utilized DeepLabCut (DLC) for accurate pig skeleton keypoint extraction and topological structure construction.
- Streamlined the Spatiotemporal Graph Convolutional Network (ST-GCN) by removing redundant layers.
- Developed an improved BCST-GCN model incorporating a global-local self-attention BC module for dynamic topological correlation reconstruction.
Main Results:
- The proposed framework effectively recognized typical pig behaviors including feeding, walking, lying, and dog-sitting posture.
- The improved BCST-GCN model demonstrated significant gains: 6.94% in accuracy, 5.61% in precision, and 6.88% in recall compared to the baseline.
Conclusions:
- The developed method offers accurate and efficient pig behavior recognition, overcoming limitations of weak temporal correlation and insufficient feature extraction.
- Provides a reliable technical solution for intelligent monitoring in pig farming, supporting the industry's intelligent upgrading.
