A dyed‐absorbance‐based framework for predicting and optimising reactive dye recipes in cotton fabric dyeing

This study proposes a data‐driven framework for predicting and optimising reactive dye recipes in cotton fabric dyeing. The proposed approach introduces dyed‐absorbance as an empirical, reflectance‐derived spectral index for wavelength‐resolved spectral representation, rather than as a direct physical absorbance. The system predicts dye combinations from dyed‐absorbance spectra and subsequently estimates the corresponding concentrations. By integrating a Convolutional Neural Network–Long Short‐Term Memory (CNN‐LSTM) model with regression‐based mathematical correction grounded in wavelength‐dependent dyed‐absorbance behaviour, the system is intended to improve colour‐matching accuracy and reduce the number of redundant dyeing trials. Unlike conventional computer colour‐matching evaluation methods that rely on reflectance‐derived K/S values and extensive empirical calibration dyeing libraries, the proposed approach is designed to reduce reliance on accumulated calibration datasets by formulating wavelength‐dependent relationships between spectral features and dye concentration. The framework was evaluated in a representative production‐derived colour‐matching case using laboratory‐scale dyeing. The CMC colour difference (Δ E ) value decreased from 1.94 to 0.43, and metamerism indices under illuminants A and U3500 were reduced to 0.29 and 0.14, respectively. The proposed framework operates on standard computing platforms and is presented as a workflow for dye recipe prediction and correction.