三维荧光光谱结合PCA-SVM对几种浓香型白酒的鉴别

发布时间：2018-04-05 21:12

  本文选题：浓香型白酒　切入点：三维荧光光谱　出处：《光谱学与光谱分析》2016年04期

【摘要】：提出一种利用三维荧光光谱技术鉴别不同品牌浓香型白酒的方法。运用FLS920荧光光谱仪测量了七个不同品牌浓香型白酒的三维荧光光谱,不同品牌浓香型白酒的荧光光谱特征相似,仅凭荧光特征参数较难区分。采用求偏导和小波压缩相结合的数据预处理方法,求解光谱数据中每一激发波长下,荧光强度对发射波长的一阶和二阶偏导数,选取db7紧支撑正交小波对数据进行压缩,选择4尺度分解后的近似系数作为新的数据矩阵,然后做主成分分析(PCA)。将提取的主成分作为支持向量机(SVM)的输入,并利用Kfold交叉验证的方法寻找支持向量机的最优参数c和γ,建立不同品牌白酒的分类鉴别模型。从每个品牌白酒中随机选取14个样本,共98个样本组成训练集,其余的42个样本组成预测集。分别比较了数据不求偏导,对数据求一阶偏导和二阶偏导的预处理后对鉴别模型的影响。结果表明:三维荧光光谱经过二阶偏导的预处理后,结合主成分分析和支持向量机能很好地实现不同品牌浓香型白酒的分类鉴别,模型的准确率为98.98%,预测集的准确率为100%。该方法具有简单,快速,成本低的优点,可为中国白酒的检测和鉴别技术的发展提供帮助。
[Abstract]:A method of identification of three-dimensional fluorescence spectroscopy of different brands of liquor. By using FLS920 fluorescence spectrometer three-dimensional fluorescence spectra of seven different brands of liquor were measured, fluorescence spectral characteristics of different brands of liquor are similar, only by fluorescence characteristic parameters is difficult to distinguish. By using the method of data preprocessing and partial derivative the combination of wavelet compression, for spectral data of each excitation wavelength, the fluorescence intensity of the emission wavelength of one order and two order partial derivative, selecting DB7 compactly supported orthogonal wavelet to compress the data, select the approximate coefficient 4 scale decomposition as a new data matrix, then principal component analysis (PCA). The extraction of principal components as support vector machine (SVM) input, and using the method of Kfold cross validation to find the optimal parameters of support vector machine C and gamma, identification and classification of the establishment of different brand of liquor Model. From each brand of liquor were randomly selected from 14 samples, a total of 98 samples consisting of the training set and the remaining 42 samples composed of prediction set. Compared the data for partial derivatives, for impact on identification model of first-order partial derivative and two order partial derivatives of the pretreatment of the data. The results showed that: three dimensional fluorescence spectra after two order partial derivatives of the pretreatment, combined with principal component analysis and support vector machine can realize the identification and classification of different brands of liquor, the accuracy of the model is 98.98%, the accuracy rate of the prediction set is 100%. the method is simple, rapid, low cost, can help to develop for the detection and identification technology Chinese liquor.

【作者单位】： 江南大学理学院;
【基金】：国家自然科学基金项目(61378037)资助
【分类号】：O657.34;TS261.7

【相似文献】

中国期刊全文数据库 前1条

1 袁瀚;易彬;沈才萍;;应用GC-MS对两种原粮酿造的浓香型白酒挥发性化合物的分析[J];酿酒科技;2014年03期

，

本文编号：1716475