Application Research In Color Matching For Textile Dyeing Of BP Neural Network Based On Clustering

Small batches,various styles and quick goods delivery are now becoming the main character of today's textile dyeing industry.Traditional color matching for textile dyeing can no longer meet to that demand due to its defects of poor precision and efficiency. Neural networks,especially BPNN(Back-propagation Neural Network) is introduced into this field and gives us a new method to solve these problems.On the other hand,long training time and low converging rate caused by BP algorithm and large scale samples have hindered the development of BPNN applied in color matching for textile dyeing.In order to enhance the performance of color matching for textile dyeing model based on BPNN,BP algorithm improving and sample retrenching through clustering are proposed in this paper.Firstly,this paper introduces some principle of neural networks and BPNN especially. Two kinds of improving BP algorithm are proposed to avoid the defects of the original BP algorithm.By comparing and analyzing,LM(Levenberg-Marquardt) algorithm is chose to construct the model of color matching for textile dyeing based on BPNN.From the theory of color difference and clustering,we get a conclusion that textile sample set belong to can-be-clustering structure under Lab color space.And the conclusion establishes the theory foundation for sample clustering and simplifying.Secondly,sample sets' collection,color space transferring,clustering analysis and standardization are proposed and finally obtain teaching samples and test samples of the BPNN model of color matching for textile dyeing based on sample clustering.Then the model is constructed step by step.Finally,train momentum BPNN and LM-BPNN separately with the clustered samples to compare their converging rate and training precision;train LM-BPNN with whole samples and simplified samples by clustering separately to compare their converging rate;train LM-BPNN with simplified samples by random select and simplified samples by clustering to compare their generalized performance.The results proved that the LM-BPNN based on simplify the samples by clustering can enhance the performance of training speed and generalization and predict dyeing recipe precisely.