Intelligent Prediction Of Fabric Dyeing Formula Base On Hyperspectral Colorimetry

Fabric color matching is a crucial technique in the printing & dyeing field.The final quality of printing & dyeing products depends on the accuracy of color matching.At present,the three stimulus values of fabrics color are mainly used to predict the dyeing recipes intelligently,which cannot solve the phenomenon of metamerism in the process of color matching.The Hyperspectral imaging System(HIS)can obtain the spectral information and location information of the fabric,which provides a possibility for the implementation of the full spectrum color matching and level dyeing evaluation of the fabric.Based on this,an intelligent prediction scheme of fabric dyeing recipe based on HIS color measurement was proposed in this thesis.The aim is to use the rich spectral information collected by the HIS to achieve accurate prediction of fabric recipe.This scheme not only avoids the problem of metamerism while using tristimulus value for fabric color matching,but also reduces the influence of uneven dyeing of fabric samples on the accuracy of formula prediction.The main contents of this thesis are as follows:1.The textile color matching theory based on HIS color measurement was studied.The double constant theory of KUBELKA-MUNK was studied,and a method of fabric color matching based on HIS color measurement was proposed.The correctness of color matching based on HIS was proved in principle,which provides a theoretical basis for recipe predicting modelling.2.The dyeing experiment was designed and completed.Dyeing experiments and HIS color measurement experiments were designed.323 level dyed samples and 20 uneven dyed samples were obtained.The selection of dyeing fabric,dyestuff,and dyeing instrument was finished and the dyeing process and the dyestuff recipe were designed.A total of 343 sample fabrics were measured on a HIS,and the reflectivity spectrum curves of the sample fabrics were obtained.3.The prediction model of fabric reflectivity-recipe based on improved deep neural network was studied.Based on the principle of deep neural network,an improved algorithm of deep neural network was proposed.A reflectivity-recipe prediction model based on this algorithm was established by using the reflectivity spectrum data of 323 level dyed fabrics.The prediction results of the test group show that the maximum absolute error and the average absolute error of the model predicted by the improved algorithm based on deep neural network are 0.02 and 0.0171 respectively.4.The prediction model of fabric reflectivity-recipe based on improved recurrent neural network was proposed and a region correlation algorithm was studied to extract the reflectance spectrum of uneven dyed fabric sample.In terms of the principle of residual network block and multi-task learning,the recurrent neural network was improved.The reflectance-recipe prediction model of was established using this algorithm based on the reflectance spectrum information of 323 leveled fabrics.A region correlation algorithm was proposed and the reflectance spectrum of uneven dyed fabrics was extracted.The improved reflectance-recip model and the region correlation algorithm were tested on the uneven dyed fabric samples.Comparisons were made between the model using the region correlation algorithm and the model not.The results show that the maximum error between the predicted recipe value and the real value is 0.004,and the average absolute error is 0.0031.