The National Institute of Standards and Technology (NIST) in the US has released NIR-SORT 2.0, expanding its spectroscopic dataset to provide high-fidelity molecular fingerprints for textile identification systems used in automated sorting. The update introduces new material specimens and instrumentation profiles to improve evaluation of classification models that interpret complex Near Infrared spectral data for determining fibre composition in collected textiles.
- Textile identification relies on Near Infrared spectroscopy, a non-invasive method that detects chemical bond structures but requires classification models to interpret complex spectral signatures.
- Earlier efforts to assess textile identification performance were constrained by limited access to high-quality, open datasets, especially for complex or proprietary textile blends encountered in sorting systems.
- NIR-SORT 1.0 addressed this gap and has supported over 400 unique users, including industry stakeholders seeking to refine their sorting algorithms.
- The Near Infrared Spectra of Origin-defined and Real-world Textiles dataset from the National Institute of Standards and Technology is available through the NIST Public Data Repository and the Materials Data Framework for model development.
DATA GAP: Validation of textile identification models has historically been constrained by limited availability of high-quality, open-access reference datasets, particularly for varied fibre compositions. NIR-SORT 2.0 builds on a curated, machine-actionable dataset to improve benchmarking of classification systems that interpret Near Infrared spectral data, addressing challenges associated with feedstock purity and blend identification in automated sorting environments.
- Many collected textiles contain complex or proprietary blends, making it difficult to establish reliable reference points for evaluating classification model performance.
- Earlier validation efforts lacked sufficiently diverse and standardised datasets, limiting the ability to benchmark machine learning systems across different textile compositions.
- NIR-SORT 1.0 partially addressed these constraints and has been used by industry stakeholders to refine textile identification algorithms.
- The expanded dataset is designed to support evaluation of classification models applied to operational textile sorting scenarios.
MATERIAL COVERAGE: Version 2.0 expands the dataset with a wider range of material specimens, experimental conditions and instrumentation profiles to improve interpretation of spectral variability across textiles. The update specifically targets feedstock purity and blend identification by introducing new fabric types, dye conditions and previously unrepresented fibre compositions, enabling more robust training and validation of models used in automated textile identification systems.
- The dataset includes 61 new in-house custom blend specimens designed to help models recognise varying fibre ratios across complex textile compositions.
- It adds 12 pre-consumer fabrics and 8 undyed fabrics featuring three previously unrepresented fibre types, broadening material diversity for model training.
- The update incorporates 12 custom-dyed fabrics to enable analysis of how colourants influence Near Infrared spectral signatures in textile identification.
- Expanded specimen diversity allows classification systems to better account for real-world variability encountered in collected textile feedstocks during sorting processes.
- The update integrates a new handheld NIR device, a standoff benchtop experiment and advanced imaging methods including polarised and colour microscopy for enhanced fabric analysis.
THE ROAD AHEAD: NIST plans to extend validation beyond digital datasets by introducing Research Grade Test Materials designed for physical testing of textile identification systems. This next phase aims to ensure that classification models perform reliably when integrated with hardware in operational environments, while ongoing dataset development and stakeholder feedback will further refine validation frameworks for industrial textile sorting applications.
- Research Grade Test Materials will provide physically characterised fabrics that partners can use to test integrated hardware and software systems under operational conditions.
- NIST is seeking contributors for a comparison study to validate these materials and support development of a standardised framework for textile feedstock identification.
- The collaborative effort aims to align testing methods and improve consistency in evaluating textile sorting technologies across different systems and environments.
- NIR-SORT 3.0 is in development and is expected to include more than 50 additional fabrics along with enhanced validation methodologies.
