Dyeing recipe prediction of cotton fabric based on hyperspectral colour measurement and an improved recurrent neural network

Abstract Precise dyeing recipe prediction is important in the final colour reproduction of textile dyeing and printing products. Currently, the widely used dyeing recipe prediction methods based on colour tri‐stimulus cannot effectively avoid the metamerism phenomenon. An intelligent dyeing recipe prediction model for cotton fabric dyeing is proposed in this paper based on hyperspectral colour measurement and a deep learning algorithm. The hyperspectral colour measurement can obtain three‐dimensional spectral information (X, Y and λ) of fabric samples, and can acquire accurate colour values even with uneven samples if the regional correlation algorithm is used. A deep learning algorithm based on an improved recurrent neural network was then employed to establish the model between spectral reflectance and the dyeing recipe. In total, 343 evenly dyed and 20 unevenly dyed fabric samples were dyed using the dyestuffs of Reactive Red CI 238, Reactive Blue CI 204 and Reactive Yellow CI 206, upon which the recipe prediction model was based, established and evaluated. The experimental results show that the proposed model based on hyperspectral colour measurement and our algorithm can provide higher prediction accuracy for Reactive Red CI 238, Reactive Blue CI 204 and Reactive Yellow CI 206. The relative prediction errors are 3.40%, 2.70% and 3.10%, respectively, for these three types of dyeing recipe, while the relative prediction errors are 19.60%, 22.60% and 11.83%, respectively, using the Datacolor 650 recipe prediction model.