Counting sheep: human experience vs. Yolo algorithm with drone to determine population

Ninahuanca Carhuas J.; Cerna L.A.; Unchupaico Payano I.; Garcia-Olarte E.; Mauricio-Ramos Y.; Quispe Euglogio C.; Hadi Mohamed M.M. · Veterinary Integrative Sciences (2025) · DOI: 10.12982/vis.2025.032

Resumen

This study assessed the efficiency of traditional human counting methods and the YOLOv7 algorithm in sheep population management at SAIS Pachacutec S.A.C., Peru. Human counters with varying experience levels (M1-M4) and the YOLOv7 algorithm (M5) were evaluated across six sheep flocks of different sizes. Traditional counting involved "linear pair counting" with human assistants, while the YOLOv7 algorithm utilized drone-captured images for automated counting. Using ANOVA and post-hoc tests, data analysis indicated that 24 months of human experience achieved 100% accuracy, highlighting the importance of expertise in accurate population management. The YOLOv7 algorithm achieved 85% accuracy, affected by factors such as the number of training images, hardware limitations, and training parameters. Despite its lower accuracy, YOLOv7 significantly reduced counting time compared to manual methods, making it a viable option for rapid object counting tasks. Further improvements in algorithm training and computational resources could enhance the algorithm's accuracy. These findings suggest that while human expertise remains critical for precise sheep counting, advancements in computer vision algorithms like YOLOv7 offer promising support, particularly for reducing counting time.

Instituciones: Universidad Nacional de Trujillo; Universidad Peruana Los Andes; Universidad Nacional del Centro del Perú