Cotton fabric moisture content quantitative detection method based on hyperspectral imaging
Xiaoke Jin,Haonan He,Jingjing Jiang,Yingjie Qiu,Xintao Qi,Huifang Zhang,Wei Tian,Chengyan Zhu
DOI: https://doi.org/10.1080/00405000.2023.2300504
IF: 1.7698
2024-01-12
Journal of the Textile Institute
Abstract:The moisture content of fabrics directly affects the physical properties of fabrics, it is important to accurately detect the moisture content of fabrics. To verify the feasibility of hyperspectral imaging technology for detecting the moisture content of cotton fabrics and thus improve the accuracy of moisture content detection, 100 groups of hyperspectral images of cotton fabrics with different thicknesses and moisture contents were acquired, the characteristics of the spectral curves of cotton fabrics were elaborated, and the effects of thickness and moisture content on the spectral curves of cotton fabrics were analyzed. Based on the research above, 1190, 1450 and 1940 nm bands were determined as the characteristic bands, the content gradient method was selected to divide the training set and sample set, PLS (Partial Least Squares) prediction models of the water content with MSC (Multiplicative Scatter Correction), SNV (Standard Normal Variate) and the first order derivative spectral pretreatment methods were established respectively, and the accuracy of the established three models was compared by correlation coefficient and standard error finally. The following conclusions were drawn: the change of water content in cotton fabric not only affects the overall reflectance of the cotton fabric spectral curve but also changes the position and peak of the characteristic peaks; with the increase of water content, the proportion of pure water in the mixed spectra of cotton fabric containing water increases, the position of the characteristic peaks is shifted and the peak increases; for cotton fabrics, the final spectral curves is the overlap of the water NIR (Near Infrared) and cotton absorption band at the position; the correlation coefficients of the training set and the test set of the established PLS-First Order Derivative model reached 0.98239 and 0.97637, the standard errors are 0.0003 and 0.00041, which proves the feasible of the method of utilizing hyperspectral imaging system with the prediction model for quantitative moisture content inspection of cotton fabrics.
materials science, textiles