A Novel Method for Estimating the Body Weight and Size of Sows Using 3D Point Cloud

Area of Science:

• Animal Science
• Computer Vision
• Agricultural Technology

Background:

• Sow body weight and size are crucial for health and reproduction.
• Manual measurement is stressful, time-consuming, and labor-intensive.
• Accurate, non-contact monitoring is needed for intensive sow production.

Purpose of the Study:

• To develop an innovative method for estimating sow body weight and size using 3D point cloud data.
• To provide an accurate and efficient tool for non-contact body condition monitoring.

Main Methods:

• Acquired 3D point cloud data using an Intel RealSense D455 camera.
• Employed a KPConv segmentation model to extract the sow's back point cloud.
• Utilized a novel dual-branch, multi-output regression model (DbmoNet) for feature integration and prediction.

Main Results:

• KPConv achieved 99.54% overall segmentation accuracy.
• DbmoNet demonstrated low mean absolute percentage errors: 3.74% for body weight, 3.97% for chest width, 3.33% for hip width, 3.82% for body length, 1.94% for chest height, and 2.43% for hip height.
• The method was validated on 2400 samples across three breeds.

Conclusions:

• The proposed 3D point cloud method accurately estimates sow body weight and size.
• This non-contact approach enhances sow health and reproductive performance monitoring in intensive farming.
• DbmoNet offers superior performance compared to existing benchmarks for sow body condition assessment.