Individual facial identification of beef and dairy cattle based on computer vision

Identifying a single animal in a herd allows for individual management association and traceability of each animal over time. Over the years, several methods have been adopted to identify animals, such as ear tags, tattoos, and electronic identification devices. However, these systems have drawbacks with one or multiple factors: durability, ease of application, cost, and welfare. Systems based on image processing can be considered alternatives for individual identification of farm animals, as they are non-invasive and easy to install. They can provide real-time information on multiple animals at a time. This work aims to analyze and develop an automatic individual identification system based on facial recognition for four different breeds of cattle (Holstein, Nelore, Angus, and Girolando). Facial images were collected from 12 animals for each breed. A YOLO v8 detection model was trained to detect animal faces and crop images. A classification model was then trained to identify animals using cropped face images as input. Results showed a mean average precision of 99% in identifying individual animals for all breeds analyzed. Therefore, the proposed model presents potential for field applicability in individual animal recognition for different breeds. Future steps of this research will include a more significant number of animals in the model and field testing to validate the methodology.