The Horse Face Individual Recognition Dataset Collected at Zhaosu Horse Farm, Xinjiang, China

With the advancement of smart ranching and intelligent livestock management, computer vision–based horse individual recognition has become increasingly important for horse identity management and information-driven breeding. However, high-quality horse face image data captured under complex natural grazing environments remain limited. To address this gap, we construct and release a horse face individual recognition image dataset tailored to natural ranch scenarios. The dataset was collected at Zhaosu Horse Farm in Xinjiang in August and November 2024, covering multiple horse breeds, with Ili horses as the predominant population. All images were acquired in unconstrained natural stable and pasture environments.During data collection, various factors were comprehensively considered, including illumination conditions, shooting distances, viewpoint variations, and differences in horse postures. Raw samples were obtained through video recording and frame extraction, followed by a series of data processing and quality control procedures, including similarity-based filtering, automatic horse face detection and cropping, and manual verification. The horse face images are organized and stored according to individual identities, with samples of each individual placed in separate folders corresponding to their identity information. The proposed dataset is characterized by high scene complexity, pronounced inter-individual differences, and strong diversity. It provides valuable benchmark data for research on horse face individual recognition, open-set recognition, cross-breed generalization, and related applications in smart ranching.