Research On Prediction Method Of Wood Dyeing Formulation Based On Improved LSTM

Using wood dyeing technology to improve the color characteristics of the veneer surface of artificial fast-growing wood is an important means to improve the utilization rate of plantation wood.In wood dyeing technology,how to quickly and accurately predict the dyeing formula is one of the key and difficult points.At present,the field of wood dyeing mostly uses color tristimulus values to predict the dyeing formula.Although this method has certain prediction accuracy,it cannot avoid the phenomenon of metamerism.In order to solve this problem,this study starts with the relationship between dyeing formula and spectral information of wood surface after dyeing,and constructs an intelligent color matching model of wood dyeing based on spectral reflectance of wood surface.The color matching model is improved from the aspects of improving the prediction accuracy of the model and reducing the structural complexity of the model.The specific research contents are as follows :(1)In order to avoid the phenomenon of metamerism in the process of color matching,according to the established data set of dyed wood,the spectral reflectance of wood surface was used as the model input,and the corresponding wood dyeing formula was used as the model output.The LSTM was used to construct the color matching model of wood dyeing based on spectral reflectance.It is optimized by batch standardization to avoid gradient disappearance.Then,from the aspect of formula prediction accuracy of the model,the LSTM color matching model is compared with two color matching models based on traditional neural networks(BP,RBF),and the superiority of the LSTM model is verified.(2)The LSTM neural network requires more parameters,which have a great influence on the prediction accuracy of the color matching model.Aiming at the problem that manual assignment is susceptible to human subjectivity and the prediction accuracy of the model may be reduced,this study proposes to use squirrel search algorithm to optimize some parameters,and use squirrel search algorithm to improve the optimized LSTM color matching model,which improves the stability and prediction accuracy of the color matching model.(3)To further avoid pseudo-random initialization,the initial population of the squirrel search algorithm was optimized using a chaotic mapping approach,and to further reduce the chance of squirrels being trapped in local minima during the operation of the algorithm,the algorithm was improved using the crossover mechanism in the difference algorithm(DE)to increase the diversity of squirrels after the squirrels finished foraging,followed by an improved location update strategy for the squirrels Finally,an improved squirrel search algorithm,ISSA,was proposed and the parameters of the model were optimised using the ISSA algorithm to obtain an ISSA-BN-LSTM model with a mean relative deviation of 0.192 for the predicted recipes of the test samples,which further improved the prediction accuracy of the wood staining recipes compared to the mean relative deviation of 0.367 for the predicted recipes of the model before the improvement.