FasterLite-YOLO: A lightweight cattle face detection model

We present FasterLite, a novel method aimed at reducing the model’s parameter count and computational requirements. This approach addresses issues such as the previous model’s excessive parameter size and high computational complexity during practical deployment and coordination with edge devices. FasterLite introduces a lightweight multi-branch network structure, replacing the embedding layer and merging layer with hidden layers. Furthermore, it combines with the C2f module in YOLOv8, resulting in the C2f-FasterLite structure. After replacing the backbone structure of YOLOv8 with the FasterLite and utilizing a dataset of cow face images captured in various complex environments, we aimed to accurately evaluate the model’s utility and robustness. Compared to the YOLOv8 model, FasterLite-YOLO exhibited a 33.44% reduction in parameters, a 33.71% decrease in floating-point operations, and achieved a detection accuracy of 98.3%. In conclusion, the experimental results demonstrate that this model effectively reduces the model’s parameter count and computational requirements, making it suitable for practical edge computing scenarios.