Quantitative Determination of the Fiber Components in Textiles by Near-Infrared Spectroscopy and Extreme Learning Machine

摘要

The quantitative determination of the components of textiles takes an important position in quality control. A rapid and green method for simultaneously determining four components in blended fabrics was explored by near-infrared (NIR) spectroscopy combined with chemometrics. Two sample sets were designed and used as the training and test sets, respectively. The four components included wool, polyester, polyacrylonitrile, and nylon. A variance sorting-based algorithm was used for variable filtering. Both classic partial least squares (PLS) and extreme learning machine (ELM) were used for multivariate calibration and a systematic comparison was made. This results reveal that ELM is superior to the conventional PLS, especially when there are fewer variables, indicating that NIR spectroscopy combined with ELM and a pertinent variable selection is feasible for NIR-based textile analysis. The developed procedure may have commercial and regulatory potential to avoid laborious, time-consuming, and expensive wet chemical analysis.