LSR-YOLO: A High-Precision, Lightweight Model for Sheep Face Recognition on the Mobile End

The accurate identification of sheep is crucial for breeding, behavioral research, food quality tracking, and disease prevention on modern farms. As a result the time-consuming, expensive, unreliable problems traditional sheep-identification methods, relevant studies have built face recognition models to recognize through facial images. However, existing such as high computational costs, large model sizes, weak practicality. In response above issues, this study proposes lightweight named LSR-YOLO. Specifically, ShuffleNetv2 module Ghost were used replace feature extraction in backbone neck YOLOv5s reduce floating-point operations per second (FLOPs) parameters. addition, coordinated attention (CA) was introduced into suppress non-critical information improve ability model. We collected images 63 small-tailed Han construct dataset further evaluate proposed method. Compared YOLOv5s, FLOPs parameters LSR-YOLO decreased by 25.5% 33.4%, respectively. achieved best performance dataset, [email protected] reached 97.8% when size only 9.5 MB. experimental results show that has significant advantages accuracy size. Finally, we integrated mobile devices developed system achieve real-time recognition. an effective method identifying sheep. fast speed, which gives it application value welfare breeding.