基于高光谱成像技术的大曲还原糖含量预测及其可视化  

Prediction and visualization of reducing sugar content in Daqu based on hyperspectral imaging technology

作　　者：刘亮 黄丹平 田建平 黄丹 罗惠波 田颖 徐佳乐 叶建秋 LIU Liang;HUANG Danping;TIAN Jianping;HUANG Dan;LUOHuibo;TIAN Ying;XU Jiale;YE Jianqiu(College of Mechanical Engineering,Sichuan University of Science and Technology,Yibin 644000,China;College of Bioengineering,Sichuan University of Science and Technology,Yibin 644000,China)

机构地区：[1]四川轻化工大学机械工程学院,四川宜宾644000 [2]四川轻化工大学生物工程学院,四川宜宾644000

出　　处：《食品与发酵工业》2022年第5期247-254,共8页Food and Fermentation Industries

基　　金：酿酒生物技术及应用四川省重点实验室项目(NJ2018-05);企事业单位委托科技项目(CXY2019ZR006)。

摘　　要：还原糖是大曲质量评价的重要指标,为进一步提高大曲还原糖含量的检测精度,提出了一种应用高光谱成像技术检测大曲还原糖含量的方法。采用高光谱成像系统,在900~1 700 nm采集大曲样本的光谱信息,并提取全部样本的平均光谱数据。首先,采用标准正态变量校正(standard normal variables, SNV)、卷积平滑、多元散射校正3种预处理方法对原始光谱进行预处理;然后,分别使用主成分分析(principal component analysis, PCA)荷载系数法、连续投影法(successive projections algorithm, SPA)和PCA荷载系数-SPA三种方法提取了大曲光谱数据的特征波段;最后,基于全波段和特征波段的光谱数据,分别建立了预测还原糖含量的偏最小二乘回归(partial least squares regression, PLSR)和最小支持向量机(least squares support vector machine, LS-SVM)模型。结果表明,对SNV预处理后的大曲光谱数据,采用PCA-SPA算法提取的特征波段建立的PLSR模型效果最佳,其中提取的特征波段数为26个,预测集相关系数为0.922 7,预测集均方根误差为0.455 6 g/100 g。基于最优的模型SNV+PCA+SPA+PLSR对不同发酵时期的大曲样本实现了还原糖含量的可视化。利用高光谱成像技术可实现大曲还原糖含量的快速检测和可视化分布,研究结果为大曲中还原糖含量的检测提供了一种新的方法。Reducing sugar is an important index for the quality evaluation of Daqu. In order to further improve the detection accuracy of reducing sugar content in Daqu, a method for detecting reducing sugar content in Daqu by hyperspectral imaging technology was proposed. The hyperspectral imaging system was used to collect the spectral information of Daqu samples in the range of 900-1 700 nm, and extract the average spectral data of all samples. Firstly, standard normal variables(SNV), smoothing convolution(SG) and multiplicative scatter correction(MSC) are used to preprocess the original spectrum. Then, the characteristic bands of Daqu spectral data are extracted by principal component analysis(PCA) load coefficient method, successive projections algorithm(SPA) and PCA load coefficient SPA respectively;Finally, based on the spectral data of full band and characteristic band, partial least squares regression(PLSR) and least squares support vector machine(LS-SVM) models for predicting reducing sugar content were established. The results showed that the PLSR model based on the feature bands extracted by PCA-SPA algorithm was the best for the Daqu spectral data preprocessed by SNV, in which the number of feature bands extracted was 26, the correlation coefficient of prediction set was 0.922 7, and the root mean square error of prediction set was 0.455 6 g/100 g. Based on the optimal model SNV+PCA+SPA+PLSR, the reducing sugar content of Daqu samples in different fermentation periods was visualized. Rapid detection and visual distribution of reducing sugar content in Daqu could be realized by hyperspectral imaging technology, which provides a new method for the detection of reducing sugar content in Daqu.

关 键 词：高光谱成像 大曲 还原糖含量 特征波长 PCA荷载系数 可视化 

分 类 号：O657.3[理学—分析化学] TS261.11[理学—化学]