Prediction and visualization of Daqu acidity based on hyperspectral imaging technology
SUN Ting1, HU Xinjun1*, TIAN Jianping1, WANG Kaizhu1, HUANG Dan2, PENG Xinghui1
1(College of Mechanical Engineering,Sichuan University of Science & Engineering,Yibin 644000,China) 2(College of Biotechnology Engineering,Sichuan University of Science & Engineering,Yibin 644000 China)
摘要 酸度值是大曲质量评价的重要指标,提出了一种基于高光谱成像技术快速检测大曲发酵过程中酸度值的方法。通过采集大曲高光谱图像并提取感兴趣区域(region of interest,ROI)的平均光谱,采用多元散射校正(multiplicative scatter correction,MSC)、标准正态变量校正(standard normal variable correction,SNV)和S-G卷积平滑后一阶导(savitzky-golay smoothing first derivative,SGFD)3种预处理方法,再通过连续投影算法(successive projection algorithm,SPA)选取最优的特征波长,分别建立偏最小二乘回归(partial least squares regression,PLSR)和最小二乘支持向量机(least squares-support vector machine,LS-SVM)预测模型,结果显示,基于SPA从SNV预处理光谱中筛选的8个最优特征波长建立的LS-SVM模型预测大曲酸度值效果最好,其中预测集决定系数(determination coefficient of prediction,R2P)为0.913 2,预测集均方根误差(root mean square error of prediction,RMSEP)为0.008 1。通过将ROI反射率输入最优的SNV+SPA+LS-SVM预测模型中,生成了大曲酸度值可视化云图,实现了不同发酵时期的酸度值及其分布的直观显示。结果表明,利用高光谱成像技术可实现大曲酸度值快速检测和可视化分布。
Abstract: Acidity value is an important index for quality evaluation of Daqu, a method based on hyperspectral imaging technology for rapid detection of the acidity value during the fermentation of Daqu was proposed. Hyperspectral images of Daqu samples and average spectrum of regions of interest (ROIs) were collected, original spectrum was pretreated by three methods including multivariate scattering correction (MSC), standard normal variable correction (SNV) and Savitzky-Golay first-order derivative (SGFD). Optimal characteristic wavelengths were selected by successive projection algorithm (SPA), then partial least squares regression (PLSR) and least squares support vector machine (LS-SVM) models were established. The results showed that the LS-SVM model built with 8 wavelengths performed better in prediction set, with determination coefficient of prediction (R2P) is 0.913 2 and root mean square error of prediction (RMSEP) is 0.008 1. By inputting the spectrum of each pixel into the optimal SNV+ SPA + LS-SVM model, the visualization of the distribution map of acidity value in Daqu was obtained, the visualization of the acidity value and its distribution in different fermentation periods was realized.
孙婷,胡新军,田建平,等. 基于高光谱成像技术的大曲酸度值预测及其可视化[J]. 食品与发酵工业, 2020, 46(17): 226-231.
SUN Ting,HU Xinjun,TIAN Jianping,et al. Prediction and visualization of Daqu acidity based on hyperspectral imaging technology[J]. Food and Fermentation Industries, 2020, 46(17): 226-231.