Related Experiment Videos
Estimation of Ewe Live Weight and Carcass Traits Using Advanced Hybrid Deep Learning and Multimodal Feature Fusion
Ahmad Shalaldeh1, Majeed Safa2, Chris Logan3
1Department of Data Science and Artificial Intelligence, Faculty of Information Technology, Al-Ahliyya Amman University, Al Salat 19111, Jordan.
Biology
|May 26, 2026
Summary
A new Vision Transformer-based Hybrid Regressor accurately predicts live weight and body composition in ewes using non-invasive imaging. This precision livestock management tool offers a scalable, interpretable solution for real-time animal measurements.
Area of Science:
- Agricultural Science
- Computer Science
- Animal Science
Background:
- Non-invasive live weight and body composition determination is crucial for precision livestock management and animal well-being.
- Traditional methods are subjective, labor-intensive, or require expensive imaging like Computed Tomography (CT).
- There is a need for accurate, scalable, and interpretable solutions for real-time livestock measurements.
Purpose of the Study:
- To propose and evaluate a novel hybrid deep learning method for predicting live weight and carcass traits in Coopworth ewes.
- To compare the performance of a Vision Transformer-based Hybrid Regressor (ViT-HR) against other deep learning models and baseline approaches.
- To ensure model interpretability and biological relevance through visual explainability techniques.
Main Methods:
- A dataset of 1184 images from 156 ewes was utilized.
- A hybrid deep learning approach was developed, including a Vision Transformer-based Hybrid Regressor (ViT-HR) incorporating Body Condition Score (BCS) and size category as auxiliary tabular variables.
- Models were compared using stringent data partitioning at the animal level, with performance validated against CT gold standards.
Main Results:
- The Vision Transformer-based Hybrid Regressor achieved a high prediction accuracy for live weight (R² = 0.93).
- Advanced fusion systems significantly outperformed baseline concatenation methods.
- Grad-CAM visual explainability confirmed the models' ability to localize relevant anatomical locations.
Conclusions:
- The developed Vision Transformer-based Hybrid Regressor provides an accurate, interpretable, and scalable solution for non-invasive live weight and body composition prediction in ewes.
- This study bridges the gap between advanced deep learning models and practical agricultural implementation.
- The findings support the adoption of AI-driven tools for enhanced precision livestock management and animal welfare.