摘要
为建立一种适用于市场在线监测玉米油掺假牡丹籽油的方法,研究了便携式785拉曼光谱技术应用于一线市场质检的潜力,制备了525个玉米油掺假牡丹籽油的油品,评估了每种混合油品便携式近红外拉曼光谱数据集的稳定性和21种混合油品拉曼光谱谱峰的差异性,分析了拉曼光谱谱峰的振动归属。随机选择16种混合油品的拉曼光谱数据集用于训练定量模型,将剩余的5种混合油品的拉曼光谱数据集用于测试定量模型。牡丹籽油与玉米油的近红外拉曼光谱存在较大差异,根据特征峰位的归属判定,拉曼光谱谱峰差异很好地反映了植物油间的脂肪酸含量差异。提出一种将长短期记忆网络结合卷积神经网络(LSTM-CNN)的模型应用于玉米油掺假牡丹籽油的量化预测,对比了偏最小二乘回归算法的预测效果。结果表明便携式785拉曼光谱的稳定性良好,具备实现对牡丹籽油掺假量的在线、快速量化的性能,且结合LSTM-CNN模型可以实现对玉米油掺假牡丹籽油含量的预测,其模型评价参数决定系数(R^(2))为0.9908,均方根误差(RMSE)为0.0299。便携式近红外拉曼光谱结合LSTM-CNN模型是一种快速、高效、可行的鉴别玉米油掺假牡丹籽油的有效方法。
To establish a method applicable for online monitoring of adulteration of corn oil with penony seed oil in the market,this paper studied the potential of portable 785 Raman spectroscopy technology for quality inspection in the first-line market and prepared 525 corn oil adulterated peony seed oil products.The stability of the portable near-infrared Raman spectroscopy data set of each kind of mixed oil and the differences of Raman spectrum peaks of 21 types of mixed oil were evaluated.The vibration attribution of Raman spectrum peaks was analyzed.The Raman spectral data sets of 16 kinds of oil mixtures were randomly selected to train the quantitative model,and the Raman spectral data sets of the remaining 5 kinds of oil mixtures were used to test the quantitative model.Significant difference existed between the near-infrared Raman spectra of peony seed oil and corn oil.According to the attribution of characteristic peaks,the difference in Raman spectra peaks could well reflect the difference in fatty acid content between vegetable oils.A model combining long short-term memory network and convolutional neural networks(LSTM-CNN)was proposed for quantitative prediction of corn oil adulterated with peony seed oil,and its prediction performance was compared with the partial least squares regression algorithm.The results show that the portable 785 nm Raman spectrometer had good stability and could achieve online and rapid quantification of peony seed oil adulteration levels.Combined with the LSTM-CNN model,it could predict the content of corn oil adulterated peony seed oil.The model evaluation parameter's coefficient of determination(R2)was 0.9908,and the root mean square error(RMSE)was 0.0299.Portable near-infrared Raman spectroscopy combined with the LSTM-CNN model is a fast,efficient,and feasible method to identify the adulteration of corn oil with peony seed oil.
作者
张凯萍
杨青波
ZHANG Kai-ping;YANG Qing-bo(School of Information Engineering,Xuchang Electrical Vocational College,Xuchang,Henan 461002,China)
出处
《粮油食品科技》
CAS
CSCD
北大核心
2024年第6期170-179,共10页
Science and Technology of Cereals,Oils and Foods
基金
河南省重点研发与推广专项(232400410357)
河南省高等教育学会高等教育研究项目(2021SXHLX190)。
关键词
牡丹籽油
玉米油
长短期记忆网络
卷积神经网络
近红外拉曼光谱
便携式拉曼光谱仪
peony seed oil
corn oil
long short-term memory
convolutional neural networks
near infrared raman spectroscopy
portable raman spectrometer
作者简介
第一作者:张凯萍,女,1983年出生,硕士,副教授,研究方向为机器学习与大数据,E-mail:kp_z@foxmail.com。
