A Posture-Constrained Infrared Thermography Framework for Dairy Cow Mastitis Detection with DAT-YOLO26

Area of Science:

• Veterinary Medicine
• Computer Vision
• Animal Health

Background:

• Bovine mastitis causes significant economic losses in dairy farming.
• Infrared thermography (IRT) offers non-invasive early detection but is hindered by cow movement.
• Head movements in cattle disrupt thermal measurements, reducing diagnostic accuracy.

Purpose of the Study:

• To develop a robust IRT-based system for early bovine mastitis detection.
• To mitigate diagnostic errors caused by cow head movements.
• To enhance the reliability of thermal imaging for udder health monitoring.

Main Methods:

• A two-stage detection framework using lateral infrared imaging.
• Pose estimation for posture classification to ensure stable sensor-to-object distance.
• Instance segmentation and an improved DAT-YOLO26 model for accurate thermal analysis.
• Testing on a 40-cow cohort (20 healthy, 20 mastitis-positive).

Main Results:

• The proposed method achieved 87.5% accuracy, 87.18% F1-score, 85% sensitivity, and 90% specificity.
• Posture consistency significantly improved IRT-based diagnostic performance.
• The DAT-YOLO26 model enhanced feature representation and detection under variable conditions.

Conclusions:

• Enforcing posture consistency is crucial for reliable IRT-based mastitis detection in dairy cattle.
• The developed framework shows promise for improving early disease diagnosis on farms.
• Further multi-farm and multi-season validation is needed for broader application.