Video Based Real Time Cattle Recognition Using Muzzle Patterns and Yolo

Abstract

Cattle recognition is a critical task in livestock management, essential for monitoring health, tracking lineage, and preventing theft. Recently used methods like face and body pattern-based recognition have challenges due to the similarity in patterns in multiple cattle, which can significantly reduce recognition accuracy. In contrast, muzzle patterns are unique and consistent across different conditions, making them a more reliable biometric for cattle recognition. To address the need for real-time recognition, we propose a method leveraging muzzle-based cattle recognition with the state-of-the-art Yolo v10 model, known for its robust and efficient detection capabilities. Our method processes muzzles to accurately identify individual cattle in a real-time environment. After training the model over 100 epochs, we observed consistent reductions in loss values, achieving near-perfect precision and recall. The model attained a best mAP50 of 1.0 and a mAP50-95 of 0.97951. These results underscore the model's high accuracy and reliability, making it an effective solution for real-time cattle recognition and management. © 2024 IEEE.