基于拉曼光谱和CNN算法的特级初榨 橄榄油的掺假量化Quantification of adulteration in extra virgin olive oil based on Raman spectroscopy and CNN algorithm
2024
乌文泽1，何凯2，吴东雷3 WU Wenze1, HE Kai2, WU Donglei3
旨在为特级初榨橄榄油掺假快速定量分析提供参考，以掺假菜籽油的特级初榨橄榄油为例，采用激光拉曼光谱实验系统获取油样的拉曼光谱数据，运用基于Inception V2结构的卷积神经网络（CNN）算法提取拉曼光谱特征并完成光谱特征与掺假量的非线性关系映射。结果表明：特级初榨橄榄油与菜籽油的拉曼光谱存在较大的差异，其中类胡萝卜素、碳碳双键、甲基和亚甲基产生的拉曼特征峰是引起差异的主要因素；所建立的CNN模型效果较好，训练集、验证集、测试集的决定系数均大于099，均方根误差均小于0.026；在低剂量掺假中，模型的预测结果仍具有一定的参考价值。综上，拉曼光谱结合基于Inception V2结构的CNN算法所建立的模型可以满足特级初榨橄榄油掺假量的快速检测。To provide a reference for the rapid quantitative analysis of adulteration in extra virgin olive oil(EVOO), taking EVOO adulterated with rapeseed oil as an example, the Raman spectral data of the oil samples were obtained using a laser Raman spectroscopy experimental system. The convolutional neural network（CNN） algorithm based on the Inception V2 structure was used to extract Raman spectral features and complete the nonlinear relationship mapping between spectral features and adulteration amount. The results showed that there was a significant difference between the Raman spectra of extra virgin olive oil and rapeseed oil, and the Raman characteristic peaks generated by carotenoids, CC, methyl, and methylene groups were the main factors causing the differences. The established CNN model performed well, with determination coefficients greater than 0.99 for the training, validation, test sets and root mean square errors less than 0.026. In low-dose adulteration, the model′s predictive performance still had a specific reference value. In summary, the model established by combining Raman spectroscopy with CNN algorithm based on the Inception V2 structure can meet the rapid detection of adulteration amount in extra virgin olive oil.
Показать больше [+] Меньше [-]Библиографическая информация
Эту запись предоставил Directory of Open Access Journals
