一种基于主成分分析和BP神经网络的卫生用纸掺杂回用纤维的甄别方法  

Identification of Hygiene Tissues Mixed with Recycled Fibers Based on Principal Component Analysis and Back-Propagation Neural Network

作　　者：黄耀驹[1] 辛丽平[1] 柴欣生[1,2] 陈春霞 陈润权 

机构地区：[1]华南理工大学制浆造纸工程国家重点实验室,广东广州510640 [2]国家纸制品质量监督检验中心,广东东莞523080

出　　处：《造纸科学与技术》2015年第1期71-77,共7页Paper Science & Technology

基　　金：国家质检总局科技计划(质量技术监督)项目(2012104018-1);教育部博士点基金项目(20110172110026)

摘　　要：由于回收纤维中含有许多有害的成分,国家有关部门已明确规定禁止使用回收纤维生产面巾纸等卫生用纸。实现对相关卫生用纸是否掺有回收纤维的科学甄别,对人体健康和安全用纸意义重大。本研究收集了68个样品并对各样品的蓝光白度、荧光白度、帚化率、抄纸方法(分为机制和手工)、残余油墨量、原生态六个指标进行实验测定,建立了主成分分析与BP-神经网络相结合的判别模型,先通过主成分分析对实验数据进行降维处理将多指标转化为少数几个保留原始数据主要信息的并且不相关的综合指标(主成分),然后将这些综合指标输入已搭建好的BP神经网络进行神经元模拟计算。研究结果表明,基于主成分分析的BP神经网络更具优越性,能有效提高网络的预测精度和预测效率,为甄别生活用纸是否含有回用纤维提供了准确稳定的方法。Because recycled fiber contains many harmful substances, the relevant state departments have banned its use in hygiene paper products. Therefore, to establish a qualitative method for identifying the presence of recycled fibers in these products is significant to human health and paper safety. This paper selected 68 samples and the information of them on six factors or parameters, i.e. , brightness, fluorescent whiteness, fiber fibrillation, effective residual ink, papermaking methods （machine - made or hand - made ） , and material sources, are collected and used for the discriminant model based on principal component analysis and back - propagation neural network. The model first reduced the dimension of the experimental data and change multi indices into the comprehensive indices that were uncorrelated and remained the information of the original data. The results showed that the PCA - BP neural network model performs better and effectively improves the accuracy and stability in the identification of recycles fiber mixed in the hygiene paper products.

关 键 词：回用纤维 主成分分析 BP神经网络 

分 类 号：TS761.6[轻工技术与工程—制浆造纸工程]