反向传播神经网络算法结合拉曼荧光光谱法定量检测特级初榨橄榄油掺假

目的 建立基于反向传播神经网络算法结合拉曼荧光光谱技术定量检测低等级橄榄油掺假特级初榨橄榄油的分析方法。方法 制备11种不同掺伪浓度的特级初榨橄榄油混合油样各10份，在相同时间、空间及目标的前提下，使用同台光谱探测系统，采集样品的拉曼光谱和荧光光谱。经过卷积神经网络去除拉曼光谱的基线，实现拉曼光谱和荧光光谱的数据预处理。根据分子光谱与电子光谱的特征差异，人为干预并设定拉曼光谱的权重，建立低等级橄榄油掺假特级初榨橄榄油的反向传播神经网络回归模型。结果 综合评估了反向传播神经网络回归模型的评价参数，特级初榨橄榄油掺假的反向传播神经网络模型的测试集决定系数为0.9716，均方根误差为0.0569，模型预测效果较好。结论 本研究提出的反向传播神经网络算法结合拉曼光谱与荧光的探测方法，满足快速检测低等级橄榄油掺假特级初榨橄榄油的定量分析需求，为评价或跟踪特级初榨橄榄油的品质提供了一种无损伤、高效率、低成本的新检测思路。

Quantitative detection of adulteration in extra virgin olive oil using back propagation neural network algorithm combined with Raman fluorescence spectroscopy

Objective To establish an analytical method based on back propagation neural network algorithm combined with Raman fluorescence spectroscopy for the quantitative detection of low-grade olive oil adulterated with extra virgin olive oil. Methods The 10 mixed oil samples of 11 kinds of different adulterated concentrations of extra virgin olive oil were prepared, and under the same time, space, and target conditions, the same spectral detection system was used to collect the Raman and fluorescence spectra of the samples. The convolutional neural network removed the baseline of the Raman spectrum, and the data preprocessing of the Raman spectrum and fluorescence spectrum was realized. Based on molecular and electronic spectra characteristics, human intervention and setting of Raman spectral weights were used to establish a backpropagation neural network regression model for low-grade olive oil adulterated with extra virgin olive oil. Results According to the evaluation parameters of the regression model, the coefficient of determination of the test set of the backpropagation neural network model of adulterated extra virgin olive oil was 0.9716, and the root-mean-square deviation was 0.0569. The model had a good prediction effect. Conclusion The backpropagation neural network algorithm proposed in this study, combined with Raman spectroscopy and fluorescence detection methods, meets the quantitative analysis requirements for rapid detection of low-grade olive oil adulteration and extra virgin olive oil. It provides a new non-destructive, efficient, and low-cost detection approach for evaluating or tracking the quality of extra virgin olive oil.