红外光谱及Bayes信息融合技术的葡萄酒鉴别研究

Application of Infrared Spectroscopy Combined with Bayesian Information Fusion for Identification of Different Original Wines

建立了将检出限不同的近红外透射光谱技术和中红外衰减全反射光谱技术进行Bayes信息融合后用于葡萄酒鉴别的方法。分别采集3种品种和3种陈酿方式的干红葡萄酒的近红外透射光谱和中红外衰减全反射光谱,用PLS-DA法分别建立基于近红外光谱和中红外光谱的判别模型,用Bayes方法实现两种判别结果的融合修正。信息融合后的结果为:葡萄酒品种鉴别模型的建模集准确率为95.08%,检验集准确率为94.68%,葡萄酒陈酿方式鉴别模型的建模集准确率为98.91%,检验集准确率为98.75%;均优于单独采用一种光谱技术的判别结果。实验表明,信息融合技术有助于模型判别效果的提高,采用Bayes信息融合技术对葡萄酒品种和陈酿方式进行快速识别是可行的。

A technique of Bayesian information fusion on near-infrared transmission （NIR） spectra and mid-infrared （MIR） attenuated total reflectance （ATR） spectra which have different limit of detection was proposed to identify different original wines. NIR and MIR spectra of three different variety wines （ cabernet sauvignon, merlot, cabernet gernischt） and different aging wines （oak barrel, oak chips, stainless steel tank） were collected separately. Partial least squares discriminate analysis （PLS-DA） method was used to establish discriminant models, and then use the Bayesian methods to achieve the integration of the two kinds of discrimination results. The recognition accuracy after Bayesian information fusion were below： for wine variety identification, the accuracy rate of cross-validation was 95.08% and validation set was 94.68% , for wine aging ways identification, the accuracy rate of cross-validation is 98.91% and validation set was 98.75%, which achieved better results on classification than individual spectroscopy. These results suggest that spectral information fusion technology helps to improve the effect of discriminant model and is feasible for fast identification on different variety and aging red wines.