NIST releases NIR-SORT 2.0 — an open dataset of fabric NIR spectra for validating fibre-content models in textile sorting; physical test materials in preparation
On 19 March 2026 the US NIST announced NIR-SORT 2.0, an expanded collection of near-infrared (NIR) spectra and microscopy images of fabrics of known composition. It is intended for training, benchmarking and validating the classification models used in automated fibre identification.
- The fibre content of collected textiles is often determined by NIR spectroscopy — fast and non-destructive, but producing complex spectra that classification models must interpret. According to NIST, validation of such models has been limited by a lack of open reference datasets; version 1.0 had served more than 400 users by March 2026.
- According to the announcement, version 2.0 adds 61 in-house custom blend specimens with varying fibre ratios, 12 pre-consumer and 8 undyed fabrics (among them three previously unrepresented fibre types) and 12 custom-dyed fabrics for studying colorant effects, plus data from a new handheld NIR device, a standoff benchtop experiment and polarised and colour microscopy images.
- The dataset record in the NIST repository (version 2.0.1, modified 28 January 2026) lists 113 fabrics: 70 of known provenance and fibre content with various dyes, 20 undyed blends, 11 post-consumer and 12 pre-consumer. Each was measured with nominally seven replicates on a benchtop instrument (833–2500 nm) and on two handheld instruments (1550–1950 nm); the spectral data comprise about 4,248 files.
- NIST plans to release Research Grade Test Materials (RGTMs) — physically characterised fabrics for testing complete sorting systems — and is seeking participants for a comparison study to validate them. NIR-SORT 3.0, with more than 50 additional fabrics, is in preparation.
Source: NIST — NIST Releases NIR-SORT 2.0: Enhancing Model Validation for Textile Feedstock (19.03.2026)