Deriving Pseudo Pure-Component NIR Spectra for Textile Blend Classification: A Methodological Framework for Researchers Working Without Pure-Fibre Reference Samples

Near-infrared hyperspectral imaging (NIR-HSI) is an established approach for optical identi
fication of textile fibre composition, with applications in automated sorting and recycling.
Apersistent limitation in this domain is the absence of pure-component reference spectra
for minority fibres — in particular elastane — in available open-access datasets. Without
pure-component anchors, standard endmember-based classification and spectral unmixing
approaches cannot be directly applied.
This framework proposes a methodology for deriving a pseudo pure-component spectral
signature for a minority fibre using only blend samples, by formulating the problem as
a linear spectral unmixing equation in which known endmembers (cotton, polyester) are
subtracted from observed blend spectra and the residual is scaled by the minority fibre’s
known composition fraction. The approach is demonstrated on elastane detection using
the OpenTextile-NIR dataset (Sormunen et al., 2026), with the derived signature validated
against NIST NIR-SORT 2.0 chemically verified spectra (Goodge et al., 2026).
The methodology is documented in sufficient generality to be replicated for any minority
fibre type on any hyperspectral dataset where pure-component spectra exist for dominant
fibres and blend composition labels are available. Statistical significance testing is embedded
at each decision point. Known limitations are indexed and propagated formally through the
derivation as error terms