Research On Color Difference Classification Of Dyed Fabrics Based On Machine Learning

With the improvement of awareness to product quality,various industries related to color quality all require high-precision color quality,which has become a kind of necessity.Especially in the field of textile printed and dyed,the color quality of finished printed and dyed products has become a strong competitiveness of enterprises in market,so enterprises regard color quality of printed and dyed products as an essential performance indicator.The initial color difference detection system of printed and dyed products was to rely on experienced professional workers to detect the color difference.Manual detection has no fixed standard and human eyes are easy to feel tired,so to some extent there is subjective consciousness,and manual detection is inefficient.With the rapid development of artificial intelligence and Internet,it has become a necessary trend for color difference detection with intelligent machine learning to replace traditional detection systems.Therefore,this paper studies the key problems of color difference detection and illumination estimation involved in color difference detection and uses machine learning to solve these problems,which is committed to building high-precision,highly stable color difference detection models and illumination estimation models.The main research contents and results of the paper are as follows:To overcome lighting problems about color difference detection,there are main two problems,the one is color difference evaluation of uneven lighting under the same standard light source,and the other is color difference evaluation under different standard light sources.This paper proposed a light estimation algorithm for printed and dyed products named regularized random vector functional link(RRVFL).There are pathological solutions due to the output weight of standard RVFL.Therefore,regularization is used to solve the problem,which constitutes a highly robust regularized RVFL illumination estimation model.Firstly,the Gray-Edge framework is used to extract the color features of sample images collected in actual scene,and the extracted color features and lighting information of sample images are constituted a data set.Then through experiments on the data set to analyze the parameters that affect the accuracy of RRVFL model,then the optimal parameter combination is selected.Finally,the model is evaluated by constructed data set,and compared with the traditional RVFL,ELM,BP,RELM,and SVR illumination estimation algorithms.Angle error,chromaticity error and T-test were used to measure performance.The experimental results show that RRVFL is the most stable and effective in predicting,when it compared with other traditional algorithms.About the mean angle error,RRVFL madethe error respectively reduce 0.00036?2.8050?3.3518?4.1669?2.9289 compared with RVFL?ELM?BP?RELM?SVR.About the mean chromaticity error,RRVFL reduced the error respectively 0.0131?0.0763?0.0232?0.0241?0.0221 compared with RVFL?ELM?BP?RELM?SVR.To improve accuracy of color difference classification in printed and dyed products,this paper proposed a novel color difference classification model,which is based on improved extreme learning machine(ELM)with differential evolution(DE)optimized whale optimization algorithm(WOA).Considering that the initial population randomly set affects the speed and quality of the WOA,so the heuristic optimization algorithm DE was used to generate a set of excellent initial populations for WOA.The WOA has a better solution direction in the beginning,so that WOA is less likely to fall into local optimum,and thus has better solving ability.Then use the WOA optimized by DE to iteratively find the best combination of input weight and hidden layer bias that affect accuracy of ELM classification,which solved the model is different every time and the algorithm is unstable because of training caused by the random initialization of the input weight and hidden layer bias in ELM.By using the improved intelligent algorithm to optimize the random parameters of ELM,a color classification model based on ELM with strong generalization ability was constructed.Firstly,the feature values representing the color difference are extracted from the fabric image under the standard light source,the color difference levels were calculated by the color difference formula,then feature values of color difference and color difference levels were constructed into a data set.After that,the experiments on collected data set are used to analyze the parameters which affect the accuracy of DE-WOA-ELM model,and the optimal parameter combination was selected.Finally,the model in this paper was evaluated through the data set,and compared with other sixteen classification algorithms.The experimental results show that the proposed DE-WOA-ELM model obtained higher classification accuracy,and has better stability and generalization ability.The mean classification accuracy of proposed method were respectively improved 2.15%,11.06%,12.11%and 0.47%,when it was compared with ELM,SVM,BP,and KELM in dataset...