基于拉曼光谱技术的桑椹花色素苷快速检测研究

花色素苷是一种天然的水溶性黄酮类色素，具有多种药用价值，广泛存在于桑椹中，成为评价桑椹产品品质的重要指标。传统检测方法费时费力，因此实现花色素苷含量的快速检测对于桑椹产品的开发利用至关重要。该研究以桑椹中的花色素苷为研究对象，探索花色素苷与拉曼光谱特性之间的关系及拉曼光谱技术对其定量检测的可行性。对桑椹及3种花色素苷标准品的拉曼光谱进行了分析，其中可将545，634和737 cm-1处的峰位作为桑椹中花色素苷的拉曼特征峰，以此判断桑椹中是否含有花色素苷，并根据其峰值的高低来定性判断花色素苷含量多少。运用多元散射校正（MSC）、基线校正（airPLS）、归一化（Normalized）三种方法及其组合方法进行光谱数据预处理，并结合PLSR筛选最佳预处理方式。比较发现最佳预处理为airPLS+MSC+Normalized，其PLSR模型效果较好，建模集决定系数为0.97，RMSE_c为2.74，预测集决定系数为0.82，RMSE_p为13.69。基于airPLS+MSC+Normalized预处理后的光谱，采用竞争性自适应重加权算法（CARS）对光谱进行特征波长筛选，将筛选出的波长变量作为输入变量分别建立了PLSR模型和SVR模型，研究两种模型的预测效果。结果表明经过CARS处理的两种模型均能对花色素苷的含量进行准确预测，其中经过CARS变量筛选建立的SVR模型效果最好，建模集决定系数为0.98，RMSE_c为1.92，预测集决定系数为0.94，RMSE_p为4.70，预测精度较高。因此拉曼光谱技术可以实现对桑椹中花色素苷含量的快速、准确预测。

Rapid Detection of Anthocyanin in Mulberry Based on Raman Spectroscopy

Anthocyanin is a natural water-soluble flavonoid pigment with various medicinal values, which is widely found in mulberry and has become an important indicator for evaluating the quality of mulberry products. Because the implementation of the traditional detection methods could cost a lot of time and effort, it is significant to achieve the rapid detection of anthocyanin content in the development and utilization of mulberry products. In this study, anthocyanin in mulberry was taken as the research object to explore the relationship between anthocyanin and Raman spectral characteristics and the feasibility of quantitative detection of anthocyanin by Raman spectroscopy. The Raman spectra of mulberry and three kinds of anthocyanin were analyzed. The peak positions at 545, 634 and 737 cm-1 could be regarded as Raman characteristic peaks of anthocyanin in mulberry, to judge whether there was anthocyanin in mulberry, and the content of anthocyanin could be qualitatively determined as per the peak values. The spectroscopic data were preprocessed with the multiplicative scatter correction (MSC), baseline correction (airPLS), Normalized and the combined methods, and the best preprocessing method was selected by combining PLSR. It could be found that the best preprocessing method was airPLS+MSC+Normalized, and the PLSR model had a better effect. In the modeling set, the coefficient of determination is 0.97 and RMSE_c is 2.74, while in the prediction set, the coefficient of determination is 0.82, and RMSE_p is 13.69. Based on the spectra preprocessed with airPLS+MSC+Normalized, competitive adaptive reweighting sampling (CARS) was adopted to extract the characteristic wavelengths of the spectra. PLSR model and SVR model were established respectively regarding the selected wavelength variables as input variables, and the research into the predicting effects of both models was conducted. As per the results, the two models processed with CARS could predict the content of anthocyanin accurately, and the SVR model established with the screening of CARS variables had the best performance in the prediction accuracy, with the coefficient of the determination being 0.98 and RMSE_c being 1.92 in the modeling set, and the coefficient of the determination being 0.94 and RMSE_p being 4.70 in the prediction set. Therefore, the rapid and accurate prediction of anthocyanin content in mulberry could be achieved by Raman spectroscopy.