基于拉曼光谱和DCGAN数据增强的珍珠粉掺伪检测研究  

Detection of Pearl Powder Adulteration Based on Raman Spectroscopy and DCGAN Data Enhancement

作　　者：谈爱玲[1] 楚振原 王晓斯 赵勇[2] TAN Ai-ling;CHU Zhen-yuan;WANG Xiao-si;ZHAO Yong(School of Information and Science Engineering,Yanshan University,the Key Laboratory for Special Fiber and Fiber Sensor of Hebei Province,Qinhuangdao 066004,China;School of Electrical Engineering,Yanshan University,the Key Laboratory of Measurement Technology and Instrumentation of Hebei Province,Qinhuangdao 066004,China)

机构地区：[1]燕山大学信息科学与工程学院,河北省特种光纤与光纤传感重点实验室,河北秦皇岛066004 [2]燕山大学电气工程学院,河北省测试计量技术及仪器重点实验室,河北秦皇岛066004

出　　处：《光谱学与光谱分析》2022年第3期769-775,共7页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划项目(2019YFC1407904);河北省自然科学基金项目(2020203010);国家自然科学基金项目(61971373)资助。

摘　　要：珍珠粉和珍珠层粉化学成分相似,但是珍珠层粉的药用价值远低于珍珠粉,并且珍珠层粉制备容易,成本底,常被不法商家用于冒充或掺入珍珠粉中流入市场,谋取利益。因此,对珍珠粉掺伪鉴别和纯度检测具有重要的意义。采用激光拉曼光谱结合深度学习研究珍珠粉掺伪快速鉴别和纯度分析。将纯珍珠粉和珍珠层粉按一定比例混合,制成珍珠粉质量百分数分别为0%,25%,50%,75%,80%,85%,90%,95%与100%共9种纯度270个模拟掺伪珍珠粉样本。然后对样本进行拉曼光谱采集,参数设置如下:分辨率为4.5 cm^(-1),积分时间为3000 ms,激光功率为20 mW。搭建了深度卷积生成式对抗神经网络(DCGAN)模型,对样本拉曼光谱进行数据增强;在此基础上,结合K近邻(K-nearest neighbor)、随机森林(random forest)、决策树(decision tree)、一维卷积神经网络(1D-CNN)4种分类器,对纯度为85%,90%,95%与100%的小比例掺伪样本进行真伪鉴别分析;同时,结合一维卷积神经网络对9种纯度的珍珠粉掺伪样本建立纯度预测的定量模型。结果表明:基于DCGAN数据增强方法所生成的拉曼光谱,与原始光谱相比,在峰值信噪比和结构相似度两个评价指标上均明显优于传统数据增强方法;在珍珠粉掺伪定性鉴别方面,DCGAN增强后的数据分别送入4种分类器,对4种小比例掺杂样本的真伪鉴别正确率均达到100%;在对9种掺伪纯度样本纯度检测方面,对测试集样本,DCGAN-1DCNN方法所建纯度定量预测模型性能最优,其决定系数R^(2)为0.9884,预测均方根误差RMSEP为0.0348,一维卷积神经网络的损失值Loss为0.0012,定量模型拟合最好。拉曼光谱结合DCGAN算法为珍珠粉掺伪鉴别及纯度检测提供一种快速简便的方法。深度卷积生成式对抗网络的数据增强方法在光谱分析技术领域具有重要的研究意义和应用价值。The chemical composition of pearl powder and nacre powder is similar,However,the medicinal value of nacre powder is far lower than that of nacre powder,and nacre powder is easier to prepare at low cost,which is often used by illegal businesses to fake or mix into nacre powder to enter the market and seek illegal interests.Therefore,identifying pearl powder adulteration and purity analysis is of great significance.This paper used Raman spectroscopy combined with deep learning technology to study the rapid identification and purity analysis of pearl powder adulteration.The pure pearl powder and nacre powder were mixed according to a certain proportion to make 270 samples with 9 kinds of purity of 0%,25%,50%,75%,80%,85%,90%,95%and 100%respectively.Then the Raman spectrum of the samples was collected,and the parameters were set as follows:the resolution is 4.5 cm^(-1),the integration time is 3000 ms,and the laser power is 20 mW.A Deep Convolutional Generative Adversarial Network(DCGAN)model was built to enhance the Raman spectra of the samples.Furthermore,K-nearest neighbor,Random forest,Decision tree,and one-dimensional convolution neural network(1DCNN)classifiers were used to identify the authenticity of a small proportion of adulterated samples with the purity of 85%,90%,95%and 100%.At the same time,a quantitative model for the purity prediction of 9 kinds of adulterated pearl powder samples was established by using a one-dimensional convolutional neural network.The results showed as follows:compared with the original spectral data,the Raman spectral data generated by the DCGAN data enhancement method was significantly better than the traditional data enhancement methods in the two evaluation indexes of peak signal-to-noise ratio and structural similarity.For the identification of pearl powder adulteration,the accuracies of the qualitative models established by DCGAN data enhancement method combing with four different classifiers have all reached 100%.For the quantitative detection of the purity of the pearl powde

关 键 词：生成式对抗网络 拉曼光谱 深度学习 珍珠粉掺伪 一维卷积神经网络 

分 类 号：O657.3[理学—分析化学]