China-UK Wheat Traits (CUWT): a dual-view field image dataset for trait-guided wheat cultivar classification

The China-UK Wheat Traits (CUWT) dataset is a dual-view, field-based wheat image and trait resource developed for trait-guided wheat cultivar classification. The dataset was collected during the 2024-2025 wheat growing season at two field trial sites: the Nanjing Agricultural University Baima trial site in Jiangsu, China, and the National Institute of Agricultural Botany Histon site in Cambridgeshire, United Kingdom. CUWT includes 80 wheat cultivars, comprising 38 cultivars from China and 42 cultivars from the United Kingdom. Images were acquired using a standardized smartphone-based imaging protocol under natural field conditions between flowering and early grain filling, when spike-level differences between cultivars are highly informative. The dataset contains paired canopy-view and lateral-view images, together with derived spike-instance data and trait information. The dataset supports the development and evaluation of Wheat Trait-guided Vision Transformer (WHT-ViT), a trait-guided deep learning framework for interpretable wheat cultivar classification. The associated trait descriptors include awn presence, spike length-width ratio, spike compactness, spike texture, spike yellowness and spikelet density. These data are intended to support research in plant phenomics, wheat cultivar classification, trait-based model interpretation and AI-assisted crop breeding. Source code, trained model weights, configuration files, testing data and processed analysis files associated with this dataset are available from the project repository: https://github.com/The-Zhou-Lab/CUWT-WHT-ViT.