Qualitative Research Of Liquor Category Based On Sparse Principal Component Analysis And SVM

In view of the disorders in the liquor industry, there is imminent to detect the true andfalse of liquor rapidly and accurately. Nowadays, infrared spectroscopy (IR) is widely usedin major areas of food security because of the prominent advantages. Thus, based on IRcombined with chemometric methods, the qualitative analysis of the liquor categoryattributes are carried out. The classification of different brands of liquor with the sameflavor and alcohol and the identification of the liquor age were studied intensively. Thework provides a powerful reference to resolve the problem of current beverage wineindustry, which is lack of authenticity control strategy to achieve rapid analysis. Finally thevintage model is used to predict and exclude the sample, which does not belong to theknown classes. The study can also be used to identify the “shoddy” technology.The main contents are as follows:(1) Use the Fourier transform infrared spectrometer to collect the IR data ofcommercial liquors, such as “Bai Yunbian”,“Yellow Crane Tower”,“Yingjia Jiaozi”,“Yanghe Lai Zhilan”,“Guan Gongfang”. And the spectroscopy of liquor is pretreated bythe techniques of excluding abnormal samples, baseline correction and standardization;(2) Facing the features of IR data with high dimension and serious overlapping,principal component analysis (PCA) and sparse principal component analysis (SPCA) areapplied to extract features and reduce dimension. Moreover, the comparison between themis analyzed and demonstrated.(3) Based on the known brands category of liquor (“Yellow Crane Tower”,“YingjiaJiaozi”,“Yanghe Lai Zhilan”,“Guan Gongfang”), two kinds of support vector machine(classics SVM and least squares SVM (LSSVM)) and Discriminant partial least squares(DPLS) as the classifier are employed to establish identification model and then treat thebrand of unknown samples. Finally, experimental results fully confirmed that the PCAcombined with LSSVM performed better than others. However, after describing andanalyzing the merit and demerit of both SVM, we decided to abandon the LSSVMclassification because of instability.(4) The identification study of liquors from six different regions (Anhui, Sichuan,Jiangsu, Hubei, Shanxi, Hebei) is developed using the combination of SPCA_SVM orPCA_SVM method respectively. The predicted results, which is according to the lineardiscriminant analysis (LDA) decision rule, show the advantages of SPCA compared withPCA. The effective features extracted by SPCA contribute to improve the classificationaccuracy. The difference of various combination ways are discussed, including LDA andSVM combined with each method of PCA, SPCA and DPLS respectively. The result ofonly misclassifying one liquor samples has highlighted the advantages of SPCA_SVM. (5) To undertake above, the combination of SPCA_SVM method is used to identifyfive vintages (12,9,6,5and3years of “Bai Yunbian”). The result of only misclassifyingfour liquor samples from SPCA_SVM also has proven the powerful advantages of SPCAand SVM in the qualitative analysis. Finally, in allusion to the new samples with53alcohol, the paper implement the strategy of probability judgment. According to set anappropriate threshold, the non-known categories of liquor samples will be excludedcorrectly.