文献类型：期刊文献

中文题名：基于神经网络的人造板装饰纸表面色泽特征分类研究

英文题名：Classification for decorative papers of wood-based panels using color and glossiness parameters in combination with neural network method

作者：李康[1] 张毛毛[1] 杨忠[1] 吕斌[1]

第一作者：李康

机构：[1]中国林业科学研究院木材工业研究所

年份：2018

卷号：3

期号：1

起止页码：16-20

中文期刊名：林业工程学报

外文期刊名：Journal of Forestry Engineering

收录：CSTPCD;;北大核心:【北大核心2017】；

基金：国家自然科学基金(31370711);国家重点研发计划项目(2016YFD0600706)

语种：中文

中文关键词：人造板装饰纸;表面色泽特征分类;色度学参数;光泽度;主成分分析;误差反向传播神经网络

外文关键词：wood-based panels decorative papers; classification of color and glossiness parameters; chroma parameters; glossiness; principal component analysis （PCA）; back-propagation （ BP） network

分类号：TS67

摘要：通过测定人造板专用装饰纸表面的色泽参数(色度学参数和光泽度),对装饰纸表面色泽特征进行量化分析,并利用色泽参数的特征信息结合误差反向传播神经网络(BP神经网络)对装饰纸进行建模分类,探讨利用装饰纸表面的色泽参数进行装饰纸表面色泽特征分类。以色泽参数数据作为神经网络的输入变量,装饰纸类型作为神经网络的输出变量,建立三层BP神经网络模型,其中,隐含层的最佳节点数为9。结果表明:通过对色泽度参数的主成分分析,增加了光泽度参数后,各类装饰纸之间的独特性增强,更利于对装饰纸进行分类。利用色度学参数(明度指数L*,红绿轴色品指数a*和黄蓝轴色品指数b*)对装饰纸进行建模分类时,判别的总正确率为80.9%,引入光泽度参数之后判别的总正确率提高至92.9%,说明利用色度学与光泽度参数结合BP神经网络可以用于装饰纸表面视觉特征的量化分析以及快速识别分类。
As a main decorative material applied to wood-based panel,there is a great demand for decorative paper.However,the quality problem of decorative paper,such as chromatic aberration,has been of a concern in the decorative paper industry. The traditional color difference evaluation method,such as artificial visual assessment,was prone to be affected by subjective factors and the efficiency of this method was relatively low. To explore the possibility of using visual parameters,which included lightness（ L＊）,index of（ red-green opponent） axis（ a＊）,index of（ yellow-blue opponent） axis（ b＊）,and glossiness（ G） to quantitatively analyze the surface characteristics of decorative papers and to classify the different types of decorative papers,and the feature information of visual parameters was extracted to establish classification model. The models of the principal component analysis（ PCA） and back-propagation（ BP）network were developed to distinguish different types of decorative papers in this study. To avoid overfitting,full-cross validation was applied for PCA modeling of samples. Based on BP network,levenberg-marquardt（ L-M） algorithm was selected for the adjustment of weights and thresholds in order to reduce goal error of parameter vector. The results showed that the classification of different types of decorative papers worked well and the total accuracy reached 80.9%when the parameters of L＊,a＊and b＊were employed. The classification accuracy increased to 92.9% when the parameter of glossiness was added,indicating that the model is accurate and practical for the quantitative analysis and rapid classification of decorative papers using the parameters of L＊,a＊,b＊and G coupled with BP network.