基于近红外及中红外光谱融合技术快速检测黄酒中的总酚含量及其抗氧化能力

为了实现对黄酒中总酚含量（TPC）及其抗氧化能力（TAC）的快速检测，探索了将傅立叶红外光谱技术应用于快速检测这两项重要指标的可行性。协同区间偏最小二乘算法（SiPLS）用于选出有效波长区间以提高模型的预测能力。支持向量机（SVM）和主成分分析（PCA）用来融合由SiPLS选出的中红外（ATR-IR）和近红外（FT-NIR）光谱的有效波段。实验结果表明基于SiPLS筛选的有效光谱变量而建立的偏最小二乘回归模型（PLS）的精度优于基于全光谱建立的经典PLS模型。基于ATR-IR建立的模型的效果略优于基于FT-NIR光谱建立的模型。此外，基于提取自ATR-IR合FT-NIR的有效区间而建立的SVM模型的预测能力要好于建立的PLS或SiPLS模型。因此，ATR-IR及FT-IR结合特征谱区筛选方法可以作为理化检测的替代手段实现对黄酒中的TAC和TPC的快速检测，同时基于两种光谱的融合技术可显著提高模型的预测精度。
In this study, Fourier-transform near infrared (FT-NIR) spectroscopy, attenuated total reflectance infrared (ATR-IR) spectroscopy and their combination for measurements of total antioxidant capacity (TAC) and total phenolic content (TPC) of Chinese rice wine (CRW) were compared. Synergy interval partial least-squares (SiPLS) algorithm was used to select wavelengths to improve PLS models and support vector machine (SVM) and principal component analysis (PCA) were applied to pre-process the merged data from two individual instruments. It was observed that models based on the efficient spectrum intervals selected by siPLS were much better than those based on the full spectra. Models from ATR-IR performed slightly better than those from FT-NIR. Moreover, SVM models based on the combination of two spectroscopies were superior to those from either FT-NIR or ATR-IR spectra, while PLS models based on the same information performed worse than those based on a single spectrum, which may be explained by some non-linearity in the data. Therefore, the integration of FT-NIR and ATR-IR was possible and could improve the prediction accuracy of TAC and TPC in Chinese rice wine