A Novel Dual-Network Approach for Real-Time Liveweight Estimation in Precision Livestock Management

Area of Science:

• Agricultural Engineering
• Computer Vision
• Animal Science

Background:

• Livestock farming automation requires efficient non-contact measurement methods.
• Existing methods often necessitate fixed animal postures or high computational loads, limiting practical application.
• There is a need for accurate and efficient liveweight estimation in unconstrained pigs for precision livestock management.

Purpose of the Study:

• To develop a novel dual-network framework for accurate non-contact liveweight estimation in pigs.
• To extract contour information from unconstrained pigs for direct liveweight calculation.
• To establish a new benchmark for accuracy and efficiency in automated livestock measurement.

Main Methods:

• A dual-network framework was designed to extract contour information from images of unconstrained pigs.
• Liveweight was estimated directly from the extracted contour data.
• The framework's accuracy and real-time performance were evaluated.
• A new dataset, Liveweight and Instance Segmentation Annotation of Pigs, was created.

Main Results:

• The developed framework achieved a high accuracy for liveweight estimation with an R² value of 0.993.
• The framework demonstrated exceptional real-time performance, reaching 1131.6 FPS when using contour information directly.
• The results indicate a significant improvement over existing non-contact measurement techniques.

Conclusions:

• The novel dual-network framework provides an accurate and highly efficient solution for non-contact liveweight estimation in pigs.
• This technology offers significant practical value for precision livestock management in real-world farming environments.
• The introduced dataset will facilitate further research and validation in automated pig monitoring.