基于FT-NIR光谱技术检测酿酒葡萄中可溶性固形物含量

Detection of Soluble Solids Content of Wine Grapes by FT-NIR Spectroscopy

应用傅里叶变换近红外光谱技术,定量分析49种酿酒葡萄的可溶性固形物含量。采用偏最小二乘回归法建立校正模型和预测模型,评价模型的预测能力和实用性。在11 987.7~3 999.5 cm-1范围,通过KennardStone算法划分样品校正集与预测集,比较10种光谱预处理方法对模型检测结果的影响。结果表明,光谱经一阶导数法处理,在11 987.7~7 498.1 cm-1和6 101.8~4 597.6 cm-1范围建模,所得定量分析模型效果最佳,分析结果精度较高,其校正相关系数（Rca）、交互验证相关系数（Rcv）、主因子数、相对标准差（RSD）和相对分析误差（RPD）分别为0.972,0.951,7,2.76%和4.03。模型经预测集检验,预测相关系数Rp达到0.961,说明建立的模型可靠,预测效果好,能满足酿酒葡萄快速、无损检测的要求。

This paper analyzed soluble solids content （SSC） of 49 wine grape varieties with FT-NIR spectroscopy, and established calibration models and predictive models by use of partial least squares regression method, furthermore evaluated the ability of predication and utility of the model. In the range of 11 987.7 cm-1~3 999.5 cm-1, calibration set and prediction set of samples were setted according to the Kennard-Stone algorithm, and then the effects of 10 kinds of spectral pretreatment methods on test results of the model were also investigated. The results showed that quantitative spectrum models, which the spectra were treated by the first order derivative spectrum, spectrum band of 11 987.7 cm-1~ 7 498.1 cm-1 and 6 101.8 cm-1-4597.6 cm-1, obtained the best analytical results. Correlation coefficient of calibration（Rca）, cross-validation correlation coefficient of determination （Rcv）, number of main factors, the relative standard deviation （RSD） and relative analytical error（RPD） of the models were 0.972, 0.951, 7, 2.76% and 4.03, respectively. The cor- relation coefficient（Rp） of predition was up to 0.961 when the models were inspected by prediction set test. It was indi- cated that the model was reliable and satisfied with requirement of fast, non-destructive testing for the wine grape.