Accurate identification of individual cattle is vital for herd management, disease control, and traceability, yet traditional methods like ear tags and RFID are labor-intensive and unreliable for large-scale use. Leveraging advances in computer vision, we propose a novel cattle recognition framework combining Vision Transformers with two-dimensional masked Retention Networks. Evaluated on a self-collected video dataset of 50 cattle, focused on muzzle features, our model efficiently handles high-resolution frames and achieves 91.5% accuracy outperforming state-of-the-art methods. The Retention Network enhances scalability by reducing computational overhead, making the system robust under challenging conditions like occlusions and variable lighting. Our approach provides a practical and high-performing solution for automated cattle identification in precision livestock farming.