An improved YOLOv8n method was used for facial recognition of Angus cattle in captive environments to solve the problem of difficulty in distinguishing facial features caused by Angus cattle's unique black fur and to achieve the accurate and non-contact recognition. A dataset containing 11 000 facial images of 200 Angus cattle at different stages of growth was constructed. Introducing an innovative and enhanced receptive field feature fusion module was introduced to enhance the model's focus on key features. A novel lightweight detection head （LPCDH） was designed for recognizing the facial feature of Angus cattle. The group Taylor pruning method was used to eliminate irrelevant neurons by estimating their importance，thereby reducing computational costs and memory usage，and improving the deployment efficiency of the model. The results showed that the improved model achieved an average recognition accuracy of 92.6%，which was 11.5，3.8，1.8，1.9，5.1，3.9，3.7，and 2.4 percentage higher that of commonly used models including SSDs YOLOv5n，YOLOv8s，YOLOv8m，YOLOv9t，YOLOv10n，RT-Detr，and Mamba-YOLO model，respectively. The designed model was improved by 3.1 percentage in 4-fold cross-validation compared with the original YOLOv8n model. It is indicated that the constructed model is optimized for lightweight memory consumption and computational requirements，making it particularly suitable for real-time recognition on mobile devices and in practical applications，significantly improving the accuracy and efficiency of recognizing the facial feature of Angus cattle. It will have immense potential in individual recognition in the livestock industry.