Pig Weight Estimation Method Based on a Framework Combining Mask R-CNN and Ensemble Regression Model.

Area of Science:

• Agricultural technology
• Computer vision
• Animal science

Background:

• Accurate pig live weight estimation is crucial for animal welfare and farm management.
• Existing methods face challenges with variable illumination and pig body posture, impacting accuracy.
• Computer vision offers a non-invasive approach to address these limitations.

Purpose of the Study:

• To develop and evaluate a computer vision system for accurate pig live weight estimation.
• To address challenges of uneven illumination and pig body bending in weight estimation.
• To compare different feature extraction and prediction strategies for optimal performance.

Main Methods:

• Utilized Mask R-CNN for precise pig contour extraction under varying light conditions.
• Employed XGBoost with actual measurements to correct for pig body bending and geometric distortions.
• Integrated corrected features and applied three combination strategies for weight prediction using Azure Kinect DK data.

Main Results:

• The XGBoost model achieved the highest prediction accuracy with MAE of 0.389, RMSE of 0.576, and R2 of 0.995.
• The Mask R-CNN + Random Forest Regressor (RFR) method demonstrated high precision across all tested strategies.
• The proposed methods significantly improved the accuracy of live weight estimation in pigs.

Conclusions:

• The developed computer vision system effectively estimates pig live weight, addressing key environmental and physical challenges.
• The combination of Mask R-CNN for segmentation and XGBoost or RFR for prediction offers a robust solution.
• This technology holds significant potential for enhancing precision livestock farming and animal welfare.