Multi‐Output Linear Regression‐Based Yarn and Fabric Spectra Prediction

ABSTRACT Fabrics from yarns of the same color but with different weaving parameters exhibit variations in their spectral properties, which are challenging to quantify description using physical models. This study proposes a bidirectional spectra prediction method for yarn and fabric based on a multi‐output linear regression (MOLR) model based on the correlation analysis between the weaving parameters and spectra. Six yarns of different colors were used to weave fabric samples with varying textures by adjusting parameters such as weft density, reed count, and weave structure on an SGA598 fully automatic rapier loom. The spectral reflectance of both yarns and fabrics was measured using an X‐Rite Color i7 spectrophotometer. Correlation analysis revealed a significant linear relationship between weft density and fabric spectral data. Therefore, a multi‐output linear regression model was developed to predict the forward and backward spectral relationship between yarn and fabric based on weaving parameters. The models were evaluated using k ‐fold cross‐validation and compared with random forest and multi‐layer perceptron models. The results demonstrated that the forward spectra prediction model (yarn to fabric) achieved an average RMSE of 0.0055 and a coefficient of determination ( R 2 ) of 0.9986, while the backward spectra prediction model (fabric to yarn) achieved an RMSE of 0.0065 and an R 2 of 0.9947. These results indicate strong consistency between the predicted and measured spectra, with the proposed model outperforming random forest (RF) and multi‐layer perceptron (MLP) models. This study provides methodological support for spectral measurement and analysis of customer fabric samples in the textile industry.