The application of deep learning techniques has become a current research hot topic in livestock farming. The facial detection of cattle enables the identification of cattle and the study of facial data (eye monitoring) to further determine the health status of the cattle. At present, the detection of cattle faces remains a challenge due to the diversity of cattle farm environments. In this study, we gathered indoor and outdoor facial data of cattle in different postures, including standing, lying, eating, and walking, and constructed a dataset consisting of 1200 cattle faces. Building on this, this paper proposes a cattle face detection model based on YOLOv11. In order to improve the model's ability to detect multi-scale information, this paper introduces the CBAM attention mechanism model. In addition, we add the GFPN network to the neck network for the detection of small target information. To further improve the generalisation ability of the model in diverse scenarios, we use ATFL to replace the traditional cross-entropy loss function in YOLOv11. The model proposed in this paper effectively detects cattle faces in various postures, both indoors and outdoors. This advancement will serve as a foundational resource for individual cattle identification, disease monitoring, and the autonomous management of cattle farms within the context of smart farming.