Related Experiment Video
Updated: May 11, 2026
09:01
Performing Vaginal Lavage, Crystal Violet Staining, and Vaginal Cytological Evaluation for Mouse Estrous Cycle Staging Identification
Published on: September 15, 2012
119.7K
Sow Estrus Detection Based on the Fusion of Vulvar Visual Features
Jianyu Fang1,2, Lu Yang1,2, Xiangfang Tang3
1Agricultural Information Institute, Chinese Academy of Agricultural Sciences, Beijing 100081, China.
Animals : an Open Access Journal From MDPI
|September 27, 2025
Summary
This study introduces an automated system for detecting sow estrus using multi-dimensional feature crossing, achieving an 85% success rate. The optimized model offers real-time processing with reduced latency and model size for large-scale pig farms.
Area of Science:
- Agricultural Engineering
- Computer Vision
- Animal Science
Background:
- Automated sow estrus detection is vital for large-scale farms to enhance reproductive efficiency and reduce labor costs.
- Conventional methods rely on subjective human expertise, leading to inaccuracies and increased husbandry expenses.
- Accurate estrus detection is critical for successful breeding and optimizing farm management.
Purpose of the Study:
- To develop an automated method for detecting sow estrus status using multi-dimensional feature crossing.
- To optimize the detection model for edge computing environments through knowledge distillation.
- To provide a reliable and efficient technical solution for real-time estrus monitoring in commercial pig facilities.
Main Methods:
- A novel method utilizing Bi-directional Feature Pyramid Network-Selective Decoding Integration (BiFPN-SDI) for feature fusion and a Spatially Enhanced Attention Module head (SEAM-Head) for signal amplification.
- Masked Generative Distillation (MGD) knowledge distillation was employed to compress the model for edge computing.
- A lightweight Multilayer Perceptron (MLP) model was developed for estrus classification based on vulvar region features and HSV color space analysis.
Main Results:
- The proposed system achieved an 85% estrus detection success rate in on-farm evaluations.
- Lightweight optimization reduced inference latency from 24.29 ms to 18.87 ms and model footprint from 32.38 MB to 3.96 MB.
- The optimized model maintained a mean Average Precision (mAP) of 0.941 with less than a 1% accuracy penalty, demonstrating robustness in complex conditions.
Conclusions:
- The developed automated system provides an efficient and reliable solution for real-time sow estrus detection in large-scale farming.
- The dual optimization strategy and lightweight compression enable effective edge computing deployment.
- This technology significantly improves breeding management and reproductive efficiency in swine production.
