The individual identification of cattle is crucial for herd management and food safety, as well as for complying with the demands of export markets, particularly those within the European Union. In addition, traditional identification methods such as ear tagging, tattooing, or hot-cold branding have significant limitations in terms of reliability, loss rates, and animal welfare. The study proposes and evaluates a non-invasive biometric identification method using the analysis of distinctive patterns in cow coat colours. The approach we use is the CLIP deep learning model (ViT-L-14) to derive a feature vector, or "biometric signature," from a picture of each cow's coat colour pattern. This method was evaluated on a large dataset (Cows2021) containing 23.350 images representing 301 unique individuals. Utilizing a cross-validation technique (80% training/20% testing), the system exhibits better performance with an accuracy of 94.28%. Additionally, performance metrics revealed precision at 94.67%, recall at 94.28%, and an F1-score at 94.27%; this result confirms the robustness of the model in the face of class imbalances. Consequently, it is believed that the extensive adoption of this method will reduce labour in herd management and improve automatic, reliable, and animal welfare-oriented identification and traceability within the livestock sector, thereby facilitating substantial advancements in precision livestock farming practices.
DEMİREL et al. (Thu,) studied this question.