激发-发射荧光矩阵光谱结合多维辩别分析用于葡萄干分类研究

葡萄干因其种类繁杂，产地来源多，制作工艺多样，导致品质各异。因此需要建立能够科学、准确的鉴别葡萄干种类、产地、品质的分析方法，以确保葡萄干产品质量、保护消费者利益、规范葡萄干商品市场。该实验基于葡萄干中富含多种荧光物质，以甲醇为萃取剂，应用微波提取法，结合三维荧光光谱技术，在激发波长300~700 nm，发射波长360~720 nm范围，获取三维荧光矩阵数据，应用多维主成分分析(M-PCA)，多维偏最小二乘辨别分析(N-PLS-DA)和平行因子算法-偏最小二乘辨别分析(PARAFAC-PLS-DA)等多维模式识别方法，对三种主色为绿色、两种主色为红色的五个不同种类的葡萄干进了分类研究。M-PCA研究结果显示不同种类的葡萄干存在聚类趋势，而N-PLS-DA和PARAFAC-PLS-DA则给出了比较满意的分类结果。相对而言，由于PARAFAC-PLS-DA是基于PARAFAC分解得到浓度得分结果基础之上进行的分类，去除了不相干的冗余信息，因此取得了100%准确的分类结果。两种算法的品质因子比较结果也说明基于荧光光谱法和多维模式识别技术相结合的分析技术可以很好的用于葡萄干种类的识别研究，并有望用于葡萄干质量等级识别及产地追溯。

Modeling Excitation-Emission Fluorescence Matrices with Multidimensional Pattern Recognition Algorithms for Classification of Raisin

With the improvement of living standards, raisins are accepted by an increasing number of people for its abundant nutrients and delicious. The quality of different kinds of raisins is very distinct because of its wide variety, diverse geographical origin, and various manufacturing technology. It is very important to establish scientific and accurate identification of variety of raisins, geographical origin and quality analysis method. These methods can not only ensure good quality of raisins and protect the consumer’s interest, but also helpful for the maintenance of the market competition order. Raisin can be measured with three-dimensional fluorescence spectrometry methods, for it contains muti-fluorescent components. In this research, fluorescence components in raisins samples were extracted with microwave method with methanol as solvent. Excitation emission spectra were obtained for 150 raisins samples of different varieties by recording emission from 300 to 700 nm with excitation in the range of 360~720 nm. The fluorescence matrix data were then analyzed by multidimensional pattern recognition methods, such as the multidimensional principal components analysis (M-PCA), multi-dimensional discrimination analysis of least squares (N-PLS-DA) and partial least square based on parallel factor algorithm discrimination analysis (PARAFAC-PLS-DA), to classify the variety of raisin. The result of M-PCA revealed the clustering tendency for the different kinds of raisins, and N-PLS-DA and PARAFAC-PLS-DA could give satisfactory classification results. In comparison, The PLS-DA classification model, constructed from PARAFAC model scores, detected the variety of raisins samples with 100% sensitivity and specificity. The study demonstrated that the excitation emission fluorescence spectrometry combining with multidimensional pattern recognition is a valuable and reliable technique for raisins classification. The results also showed that this method is promising to discriminate the quality and trace the geographical origin of raisins.