化学计量学结合中红外光谱的浓香型白酒分类研究  

Classification of strong-flavor Baijiu based on chemometrics and mid-infrared spectroscopy

作　　者：周瑞 陈晓明[1] 张莉丽 张良[2] 许德富[2] 张宿义[2] 代小雪[2] 毛洪川 谢菲 代汉聪[2] 宋艳[2] 郭佳[2] 陈雯月 ZHOU Rui;CHEN Xiaoming;ZHANG Lili;ZHANG Liang;XU Defu;ZHANG Suyi;DAI Xiaoxue;MAO Hongchuan;XIE Fei;DAI Hancong;SONG Yan;GUO Jia;CHEN Wenyue(School of Life Science and Engineering,Southwest University of Science and Technology,Mianyang 621000,China;Luzhou Laojiao Co.Ltd.,Luzhou 646000,China;Sichuan Mianyang Fenggu Wine Industry Co.Ltd.,Mianyang 621000,China)

机构地区：[1]西南科技大学生命科学与工程学院,四川绵阳621000 [2]泸州老窖股份有限公司,四川泸州646000 [3]四川省绵阳市丰谷酒业有限责任公司,四川绵阳621000

出　　处：《食品与发酵工业》2023年第5期88-93,共6页Food and Fermentation Industries

基　　金：四川省重大科技专项(2019ZDZX0003)。

摘　　要：为探究无损、快速和准确判别不同类型浓香型白酒的方法,该研究选取不同酒度和不同品牌浓香型白酒作为研究对象。利用傅里叶变换中红外光谱仪采集120份酒样的原始光谱,结合平滑滤波与标准正态变换分别对原始光谱进行预处理,采用主成分分析比较光谱预处理效果。光谱数据按Kennard-Stone方法以7∶3的比例划分为训练集和测试集,经数据归一化后使用蚱蜢算法优化支持向量机和误差反向传播人工神经网络进行建模分析。试验结果表明,光谱预处理结合主成分分析不能区分不同酒度和品牌的浓香型白酒,但平滑滤波处理后不同酒度酒样的聚类区分较好,标准正态变换处理后不同品牌酒样的聚类区分更好,二者都能有效减少中红外光谱的噪音,提高识别的精度。基于蚱蜢算法优化支持向量机和误差反向传播人工神经网络模型进行判别时,训练集和测试集的酒样分类准确率均为100%。综上所述,利用中红外光谱结合化学计量学可快速准确地判别不同酒度以及不同品牌浓香型白酒,可为白酒的香型区分、产地溯源、市场监管和售后管理等提供数字化方案。In order to explore the non-destructive, rapid and accurate method of distinguishing different types of strong-flavor Baijiu, different alcoholic strength and different brands of strong-flavor Baijiu were selected as the research object in this study. Using the Fourier transform mid-infrared spectrometer to collect the original spectra of 120 Baijiu samples, combining the smoothing filtering and the standard normal variate method to preprocess the original spectra respectively, and the principal component analysis was used to compare the spectral preprocessing effects. The spectral data were divided into training set and test set according to the Kennard-Stone method with a ratio of 7∶3. After the data normalized, the grasshopper algorithm was used to optimize the support vector machine and the error back-propagation artificial neural network for modeling and analysis. The test results showed that spectral preprocessing combined with principal component analysis cannot distinguish strong-flavor Baijiu with different alcoholic strength and brands, but the clustering distinction of Baijiu samples with different alcoholic strength after smoothing filtering treatment was better, and the clustering distinction of different brands of Baijiu samples after standard normal variate processing was better, both of them can effectively reduce the noise of mid-infrared spectrum and improve the recognition accuracy. When the discriminant analysis was performed based on the grasshopper algorithm was used to optimize the support vector machine and the error back-propagation artificial neural network models, the classification accuracy of Baijiu samples in both the training set and the test set was 100%. In summary, the method of mid-infrared spectroscopy combined with chemometrics can identify strong-flavor Baijiu with different alcoholic strength and brands quickly and accurately, and can provide digital solutions for Baijiu aroma differentiation, origin traceability, market supervision and after-sales management.

关 键 词：化学计量学 中红外光谱 浓香型白酒 分类 支持向量机 神经网络 

分 类 号：TS262.31[轻工技术与工程—发酵工程] O657.33[轻工技术与工程—食品科学与工程]