HyperTex: An Annotated Hyperspectral Dataset for Textile Fibre Composition Estimation

HyperTex is a curated hyperspectral imaging dataset of textile materials developed to support research in textile fibre composition estimation. The dataset was assembled in the context of textile recycling and circular economy initiatives, where accurate identification of fibre content is essential for automated sorting and material recovery processes.

The dataset contains 198 hyperspectral captures acquired using a Specim FX17 camera. The samples represent a diverse collection of textile materials, including both pure fibres and fibre blends commonly found in commercial textile products. Most samples were validated through laboratory chemical analysis, while the remaining samples were labelled according to manufacturer, vendor, or retail composition information.

Each sample is identified by a letter prefix (F, M, T, MC, or X) followed by a numerical index. These prefixes primarily indicate different acquisition batches and collection campaigns. The dataset includes raw hyperspectral measurements, calibration data, reflectance-normalized data, acquisition metadata, and pixel-level composition annotations. In addition, ground-truth composition maps are provided to facilitate the development of supervised machine learning methods.

The annotations were generated at pixel level, enabling the investigation of fibre composition estimation from hyperspectral imagery as both a regression and classification problem. The dataset is intended for research in hyperspectral imaging, machine learning, computer vision, and textile material characterization.

Detailed documentation, software tools, and standardized data partitions used in the associated publication are provided separately through the accompanying repository and supplementary datasets.