A Pig Mass Estimation Model Based on Deep Learning without Constraint.

Area of Science:

• Animal Science
• Computer Vision
• Machine Learning

Background:

• Accurate pig body mass estimation is crucial for monitoring growth, health, and welfare.
• Current contactless methods require pig restraint, posing challenges for animal welfare and safety.
• A need exists for non-invasive body mass estimation in unconstrained environments.

Purpose of the Study:

• To develop and evaluate a deep learning model for estimating pig body mass in unconstrained settings.
• To improve the accuracy and efficiency of contactless pig body mass estimation.
• To provide a tool for real-time assessment of pig body quality for breeding management.

Main Methods:

• Developed a deep learning model integrating Mask R-CNN for instance segmentation and Keypoint R-CNN for keypoint detection.
• Employed an improved ResNet backbone with multi-branch convolution, depthwise convolution, and inverted bottleneck for enhanced accuracy.
• Constructed a dataset comprising images and body mass data from 117 pigs.

Main Results:

• The developed model achieved a Root Mean Square Error (RMSE) of 3.52 kg on the test set.
• Outperformed existing pig body mass estimation algorithms using ResNet and ConvNeXt backbones.
• Demonstrated a high estimation speed of 0.339 seconds per frame.

Conclusions:

• The deep learning model enables accurate and efficient contactless pig body mass estimation in unconstrained environments.
• This technology offers significant potential for real-time body quality assessment, aiding in breeding plan adjustments.
• The model contributes to improved animal welfare and farm safety by eliminating the need for pig restraint.