提出基于四元数主成分分析的三维荧光光谱特征提取新方法, 并将其运用于品牌食醋溯源研究。 首先利用F7000荧光光谱仪测得不同品牌食醋样本的三维荧光光谱数据, 获取样本的等高线图和三维投影图, 并进行三维荧光等高线图分析; 然后利用激发波长分别为380, 360和400 nm下的发射光谱数据建立食醋三维荧光光谱数据的四元数并行表示模型, 对四元数荧光光谱矩阵进行四元数主成分特征提取, 并基于乘积运算、 模值运算和求和运算三种方法对提取出来的四元数主成分特征进行特征融合; 最后将融合特征作为K近邻分类器的输入, 得到不同食醋品牌的最优分类模型。 分别讨论三种不同特征融合方法和四元数主成分个数与最终模型分类正确率之间的关系。 针对四个不同食醋品牌120个样本的分析结果可得: 基于求和特征融合运算所得到的融合特征可以利用最少的特征数目, 建立最优的溯源模型, 样本预测集溯源正确率可达100%。 研究结果表明: 四元数主成分特征提取和特征融合方法能够并行表示三维荧光光谱数据所蕴含的丰富信息, 为三维荧光光谱数据分析提供新思路。

A new method was put forward to study vinegar brand traceability based onthree-dimensional fluorescence spectra technology combined with quaternion principal component analysis. Firstly, the three-dimensional fluorescence spectral data of vinegar samples with different brands were acquired by F7000 fluorescence spectrometer. The contour and 3D fluorescence spectra about four different brands vinegar were acquired and the three-dimensional fluorescence contour maps were analyzed; Then the parallel quaternion matrix representation model of vinegar three-dimensional fluorescence spectral data was established by using the emission spectral data under excitation wavelength of 380, 360 and 400 nm respectively. The quaternion features were extracted using quaternion principal component analysis, and the exacted quaternion principal components were conducted feature fusion based on operations of multiplication, modulus and summation respectively; At last, the fusion features were as the input of K-Nearest Neighbors, and the optimal classification model of vinegar brand traceability was made. The relationships between the model classification accuracy and the three different feature fusion methods and the number of quaternion principal components were discussed. According to the analysis results with 120 vinegar samples of four different brands, the fusion feature obtained by summation operation can establish the best traceability model by using the least number of features, and the accuracy of the prediction set can reach 100%. The results of this study showed that the quaternion principal component feature extraction and feature fusion methods can represent the rich information contained in the three-dimensional fluorescence spectral data in parallel, which provides a new idea for the analysis of three -dimensional fluorescence spectral data.