Neuronal network algorithm towards color model of full color gamut yarn and its color prediction research

Abstract

This article builds a gridded color matching model based on digital color mixing of four primary colored fibers in the context of digital spinning, which can separately perform regulation of hue, saturation, and lightness within the full color gamut. Combining the requirements of a neuronal network prediction algorithm, the mixing ratios of primary colored fibers and spectral reflectance were used as input and output parameters for each other, and 41 grid point mixing samples were selected from the constructed gridded color mixing model, 31 of which were chosen as training samples to establish a neuronal network model containing input, hidden, and output layers. The remaining 10 mixed samples were then selected as predicting samples to perform validation of the model's capability to predict the color or the mixing ratio of the four primary colored fibers. The final results showed that the color difference of the mixed samples used for color prediction was 0.90 at the minimum, 2.31 at the maximum, and 1.45 at the average; the mean absolute error for proportional forecasting was 0.19%, and the root mean square error was 0.97%. The findings indicate that the forecasting method has excellent prediction accuracy, and the constructed color matching model of four primary colors and neuronal network algorithm can be applied to the prediction of mixed colors of multiple primary colored fibers, providing technical support for the spinning of colored yarns.