蜂王浆是一种具有抗氧化、 抗衰老、 调节心血管系统和免疫功能的纯天然营养保健食品, 近年来在食品、 生物医学等领域广泛应用。 由于蜂王浆的采集过程费时费力且没有快捷简便的方法检测其品质, 使得市场上的蜂王浆产品质量参差不齐, 因此实现蜂王浆品质的快速鉴别就显得至关重要。 该研究以蜂王浆的水分和蛋白质为研究对象, 利用拉曼光谱技术结合主成分回归算法(PCR)和偏最小二乘法对蜂王浆进行了快速定量检测, 建立了水分、 蛋白质的定量模型, 探究对其定量分析的可行性, 并进行光谱预处理以提升模型的预测能力, 使其预测准确性更高。 蜂王浆中水分和蛋白质化学值的测定分别采纳蜂王浆国家标准规定的减压干燥法和凯氏定氮法。 蜂王浆光谱的采集则是由DXR激光共焦显微拉曼光谱仪测得。 应用TQ Analyst分析软件对蜂王浆光谱进行预处理及建立定量分析模型。 其中光谱预处理包括导数、 标准正态变换、 多元散射校正、 Savitsky-Golay卷积这四种光谱预处理法, 并按一定关系排列组合成多种不同的预处理方法, 对蜂王浆样品光谱进行数据处理, 寻找出最优的模型与处理方法。 结果表明, 利用主成分回归法建立蜂王浆水分和蛋白质的定量模型效果不理想, 水分的定量模型结果表明, Savitsky-Golay平滑(7) 处理校正集决定系数最高但也仅为0.741 3, 预测集决定系数为0.661 6, RMSEC为0.656, RMSEP为1.34, 建模效果差。 蛋白质的PCR定量模型结果表明, Savitsky-Golay平滑(7) 处理相较之下最优, 校正集决定系数0.675 0, 预测集决定系数为0.566 8, RMSEC为0.548, RMSEP为0.957, 建模效果较差。 因此, 基于PCR所建模型对蜂王浆水分、 蛋白质的含量有一定的预测可能性, 但建模效果较差, 预测准确度低, 稳健性差。 而结合偏最小二乘法并进行S-G(7) +二阶导数+SNV处理对蜂王浆水分建模效果最好, 水分含量校正集和预测集的决定系数分别为0.992 7和0.948 8, RMSEC和RMSEP分别为0.162和0.442。 蛋白质的PLS定量模型, 通过对多种预处理组合处理结果进行对比, S-G(7) +一阶导数+SNV处理对蜂王浆蛋白质建模效果最佳, 蛋白质含量校正集和预测集的决定系数分别为0.991 6和0.879 5, RMSEC和RMSEP分别为0.143和0.497, 建模效果好。 因此, 利用拉曼光谱结合偏最小二乘法快速检测蜂王浆中水分和蛋白质的含量是可行的, 且所建定量模型稳健性良好, 预测准确度高。 通过上述实验可总结得出, 在一些不可避免的外界因素影响下, 将多种预处理方法组合起来可以提高模型的准确性和稳健性, 比用单一的光谱预处理方法修正光谱更加有效, 优化效果更加明显, 且有效提升了模型的各参数, 更好的提高了模型预测的准确性。 同时表明了, 拉曼光谱技术应用于蜂王浆品质的快速检测是可行的, 且检测准确度高, 速度快, 在蜂王浆品质的快速检测方面展现了很好地应用前景。

Royal jelly is a natural nutrient health food that has antioxidant, anti-aging, regulate cardiovascular system and immune function. In recent years, royal jelly has been widely applied in food, biomedicine and other fields. Because the collection process of royal jelly is time-consuming and the quality of the royal jelly is difficult to detect and the quality of royal jelly is uneven in the market. It is very important to realize the rapid identification of the quality of royal jelly. Therefore, the content of two components of moisture and protein on the quality of royal jelly is explored in this paper, and the quantitative analysis model of principal component regression (PCR) and partial least squares (PLS) method is established for the moisture and proteir of royal jelly by Raman spectroscopy, and the feasibility of the quantitative analysis of royal jelly is explored. In the experiment, the determination of the moisture and protein chemical vollues in royal jelly adopts the vacuum drying method and kjeldahl method as specified in the national standard of royal jelly. The royal jelly spectrum is collected by the DXR laser confocal microscopy Raman spectrometer. The TQ Analyst analysis software is used to pretreat the full spectrum of royal jelly and establish a quantitative analysis model. Among them, four spectral preprocessing methods includes first derivative, second derivative, Standant Normal Variate Transformation, Multipicative Scatter Correction and Savitsky-Golay, convolution these four spectral preprocessing methods are combined into a variety of different pretreatment methods. A lot of experiments have been carried out on royal jelly samples to find out the best models and treatment methods. The results show that the effect of the quantitative model by using the principal regression method to establish moisture and protein of royal jelly is not ideal. The results of the quantitative model of moisture show that the best spectral processing method of PCR is Savitsky-Golay smoothing (7), but it is only 0.741 3. The coefficient of prediction set is 0.661 6, RMSEC is 0.656, RMSEP is 1.34, and the modeling effect is not ideal. The quantitative model of protein show that the best spectral processing method of PCR is Savitsky-Golay smoothing (7), the coefficient of correction set is 0.675 0, the coefficient of prediction set is 0.566 8, RMSEC is 0.548, RMSEP is 0.957, and the modeling effect is bad. Therefore, the model based on PCR have a certain prediction possibility for the content of moisture and protein in royal jelly, but the modeling effect is poor, the prediction accuracy is low, and the robustness is poor. The PLS method is used to establish quantitative model of moisture and protein, and S-G(7)+second derivative +SNV is the best spectral processing method for moisture of royal jelly, the coefficient of correction set and prediction set are 0.992 7 and 0.948 8, RMSEC and RMSEP are 0.162 and 0.442 respectively. And S-G (7)+first derivative +SNV is the best spectral processing method for protein of royal jelly, and the coefficients of correction set and prediction set are 0.991 6 and 0.879 5, respectively, RMSEC and RMSEP are 0.143 and 0.497, respectively. The modeling effect is ideal. The results show that it is feasible to detect moisture and protein content in royal jelly by Raman spectroscopy combined with partial least squares, and the established quantitative model has good robustness and high prediction accuracy. Through the above experiments, we can conclude that under the influence of unavoidable external factors, the combination of various pretreatment methods can improve the accuracy and robustness of the model. It is more effective than single spectral pretreatment method to correct spectra, and the optimization effect is more obvious. It effectively improves the parameters of the model, and improves the accuracy of model prediction. It is also shown that Raman spectroscopy is feasible for rapid detection of royal jelly quality, and it has high accuracy and speed and has great prospects in the rapid detection of royal jelly quality.