Sheep identification based on face features with the method and practical application analysis
Hongbo Yuan, Zhaohan Liu, Zhenjiang Cai, Yingjie Zhang, Man Cheng · Smart Agricultural Technology · 2025
Individual animal identification is foundation to precision livestock farming under intensive production systems. To avoid the stress and injury caused by invasive ear tags, we developed a deep learning model, SheepFaceRec, for individual sheep recognition using facial images. The SheepFaceRec model uses the MTCNN module to detect and segment the sheep faces, and the FaceNet module to extract facial features and generate an identity feature file, which is then compared with existing entries in the database to determine identity. To evaluate SheepFaceRec’s practical performance, we conducted experiments with 564 sheep and a generalization test with 50 additional sheep, and analyzed the impact of facial rotation angles on identification accuracy. Results showed that SheepFaceRec achieved 97.14% accuracy on the test set of 564 sheep and 95.88% on the generalization test set of 50 sheep. Accuracy remained 94.71% when the facial rotation angle was within 30°. These findings suggest that SheepFaceRec can be effectively applied in real-world farming environments for individual sheep identification.