CAT-CBAM-Net: An Automatic Scoring Method for Sow Body Condition Based on CNN and Transformer

Area of Science:

• Animal Science
• Computer Science
• Artificial Intelligence

Background:

• Sow body condition scoring is crucial for optimizing nutrition and reproductive performance in swine farming.
• Manual assessment methods are labor-intensive and time-consuming, particularly in large-scale operations.
• Accurate body condition scoring directly impacts sow health and productivity.

Purpose of the Study:

• To develop an automated system for sow body condition scoring using artificial intelligence.
• To improve the efficiency and accuracy of sow body condition assessment in commercial swine farms.
• To enhance the capture of both local and global features in sow images for better analysis.

Main Methods:

• A dual neural network architecture combining Convolutional Neural Networks (CNNs) and transformer networks was employed.
• A Channel-Attention module (CBAM) was integrated to focus on relevant image features.
• An optimized focal loss function was utilized to address data imbalance and mislabeling issues.

Main Results:

• The developed method achieved high performance metrics: 91.06% average precision, 91.58% average recall, and 91.31% average F1 score.
• Comparative experiments confirmed the proposed method's superior performance on the dataset.
• The system demonstrated effective automatic scoring of sow body condition.

Conclusions:

• The AI-powered system offers a significant advancement over manual sow body condition scoring.
• This technology has promising applications for improving efficiency and accuracy in swine management.
• Automated scoring can lead to better-informed nutritional decisions and enhanced sow reproductive outcomes.