Bidirectional Prediction Of Fabric Structural Parameters And Style Score Based On Integrated Neural Network

With the progress of the times,people’s pursuit of clothing is no longer only about style,but also the feel and style of clothing fabric.Fabric designers hope to quickly understand the style of the fabric through the structural parameters of the fabric,so as to simplify the design steps.In this paper,by comparing the two-way predictability of fabric structure parameters and fabric subjective style by different neural network models,an integrated neural network model based on data fusion theory is proposed.In this paper,the structure parameters such as yarn fineness,fabric density,weight,width and texture,as well as the subjective style scores of main tone,texture,pattern,drape,smooth,luster,softness and transparency are selected as the input and output of neural network.Through data cleaning,unified dimension and data normalization,the above indexes are suitable for neural network operation.In this paper,radial basis function(RBF)neural network and back propagation(BP)neural network are used to predict fabric samples.Because two different neural networks can produce different prediction results for the same fabric,this paper proposes a D-S model based on the arithmetic average method of data fusion theory,a weighted average combination model of evidence correlation coefficients and a weighted average combination model of error variance.The fusion of the data predicted by the two neural networks effectively improves the accuracy and precision of the prediction model.This paper studies the integrated neural network that uses fabric structure parameter clustering and classification based on fiber content composition as a classifier.It is verified using 556 fabrics from China International Textile Fabrics and Accessories Fair.When the structural parameters predict the subjective style,The accuracy of the integrated neural network using the clustering classifier is up to 9.12% higher than that of the single neural network.The accuracy of the integrated neural network using the fiber component classifier is up to 21.16% higher than that of the single neural network.