Research On Application For Color Matching In Textile Dyeing Based On BP Neural Network Adapted By Bayesian Regularization Algorithm

The traditional color matching methods are carried out by the experience and the knowledge of the workers themselves,so it brings some disadvantages.Such as hard work, materials and time wasting.In this paper,in order to overcome the disadvantages of the traditional methods,the neural network technology based on Bayesian is introduced into the field of textile color matching.A network model which has both high precision and high generalization is established,due to the neural network which can solve non-linear problems.Firstly,the knowledge about the computer color matching and the chroma of color are introduced in this paper.Due to the advantage of the computer color matching,the neural network technology is introduced into the field of textile color matching.Secondly, it introduces the knowledge about the neural network,including its theory,character and the research status.Some typical network models are also introduced in this paper.Subsequently,the knowledge about BP network and RBF network are mostly introduced.According to the respective character of the two networks,network models are established based on different methods.Aiming at the searching character of GA and combining the RBF network,a RBF network model based on genetic algorithm is established;Aiming at the traditional BP algorithm has slow convergence speed,but LM algorithm has high precision,a BP network model based on joining the Bayesian algorithm and the LM algorithm is established.The network model can not only solve the problem of convergence but also improve generalization of the network.The algorithm mainly modifies the function of error.Thereby it can improve the generalization of network.In this paper,it mainly researches how to improve the generalization of network.The experiment results show that BP network model based on joining the Bayesian algorithm and the LM algorithm can improve the generalization of network preferably.Finally,using the sample of the dark,medium and light color pigments that we selected from the factory,we make use of the GA-RBF,the join of the Bayesian and the LM,LM algorithm and the algorithm of adapting learning rate to make the simulation experiment respectively.Comparisons of simulation error and generalization,the results of the experiment show that the neural network based on the algorithm joining the Bayesian and LM has improved not only in the simulation precision but also in the generalization.