Colour transfer framework and modular back‐propagation neural network prediction system constructed for full‐colour‐gamut colour spinning

The nonlinear colour transfer effect caused by the three‐dimensional spatial mixing of coloured fibres during the processing of coloured spun yarn makes it difficult for traditional linear models to accurately describe the mapping relationship between the recipe and the resulting colour. This has become a bottleneck restricting the digital development of the coloured spun yarn industry. To address this issue, a colour transfer framework consisting of the PCS shared colour space, a standard colour model, a full‐colour‐gamut grid mixing model, a spinning twin model and a colour mixing chromatogram model is constructed in this paper. Digital characterisation of the discrete mapping of nonlinear colour transfer within this framework is provided through the constructed modular back‐propagation neural network (MBPNN) prediction system. First, using eight primary‐colour fibres (red, yellow, green, cyan, blue, magenta, white and black) as raw materials, a full‐colour‐gamut parametric grid mixing model with the same topology as the Ostwald colour solid is constructed. This model includes six colour matching sub‐models: red‐yellow‐white‐black, yellow‐green‐white‐black, green‐cyan‐white‐black, cyan‐blue‐white‐black, blue‐magenta‐white‐black and magenta‐red‐white‐black. Furthermore, a BP neural network algorithm is introduced, and a full‐colour‐gamut MBPNN neural network prediction system consisting of an input layer module, a hidden layer module, an output layer module, a database module and a prediction algorithm module is designed. This system can achieve both forward prediction (mixing concentration → spectral reflectance) and reverse prediction (spectral reflectance → mixing concentration). Finally, 218 training sample blended yarns (for model training and internal evaluation) and 36 independent prediction sample blended yarns (for external generalisation validation) are prepared using a three‐channel numerical control rotor spinning machine. The results show that in the external validation, the forward prediction achieves an average CMC(2:1) colour difference of 0.939 with a qualification rate (≤1.5) of 86.1%; the reverse prediction achieves an average colour difference of 0.848 with a qualification rate of 88.9%. Both values are below the industrial tolerance threshold of 1.5. This validates the feasibility of the full‐colour‐gamut colour transfer framework and the MBPNN prediction system constructed in this study, achieving bidirectional conversion between primary‐colour mixing ratio and colour value, thereby providing a digital characterisation tool for the nonlinear colour transfer in full‐colour‐gamut rotor colour mixing spinning.