基于紫外拉曼光谱的转基因大豆油快速识别方法研究

转基因技术对实现作物增产增质，降低农药使用量，降低生产成本等具有重要作用，但对生态环境也存在一定的潜在威胁。为了防止转基因大豆在食品化中的滥用，对转基因产品快速鉴别技术的研究尤为迫切。紫外拉曼光谱检测技术具备外场远距离无损遥测检测，简单高效，快速准确等优点，可有效用于物质遥测鉴别领域。基于紫外拉曼光谱的转基因/非转基因大豆油以及与其他类别食用油鉴别方法，采集了五种不同食用油(两种品牌转基因/非转基因大豆油各500组样本和一种稻米油100组样本，共2 100组样本)在3 500~400 cm-1(268~293 nm)范围内的日盲紫外拉曼光谱信息，为提高光谱数据的信噪比并保证分类识别的准确性，对上述光谱数据采用Savitzky-Golay滤波降噪、基于自适应迭代加权惩罚最小二乘法(airPLS)的基线校正以及多元散射校正(MSC)的光谱数据修正等预处理。根据大豆油的紫外拉曼指纹图谱，分析出主要化学成分包含脂肪类、蛋白质类、酰胺类。将每种大豆油样本按1∶1划分为训练集和测试集，输入训练集数据至支持向量机(SVM)进行训练，采用10折交叉验证建立最佳模型，识别准确率达99.81%，对转基因大豆油的判别效果显著；采用主成分分析法(PCA)进行数据降维处理，提取出8个主成分，累计贡献率为74.84%，可代表大部分原始数据特征。在此基础上，将预处理后的光谱数据按4∶1划分为训练集和测试集，采用偏最小二乘回归判别分析方法(PLS-DA)，结合10折交叉验证法建立全谱的最佳PLS-DA模型(判别阈值设置为0.5)，判别准确率达到70.95%。研究表明，紫外拉曼光谱分析方法可较为准确地鉴别非转基因/转基因大豆油，同时可鉴别大豆油与稻米油，实现对转基因大豆食品的快速无损鉴别，可望成为转基因大豆油及其食品的现场检测新的技术途径，对推动转基因产品遥测鉴别技术的发展具有进步意义。

Rapid Identification of Transgenic Soybean Oil Based on Ultraviolet Raman Spectroscopy

Transgenic technology plays an important role in increasing crop yield and quality, and reducing pesticide use and production cost, but it also has a certain potential threat to the ecological environment. In order to prevent the abuse of genetically modified soybean in food, the research on rapid identification technology of genetically modified products is particularly urgent. UV Raman spectroscopy detection technology can be effectively used in material telemetry and identification with many advantages, such as long-distance non-destructive telemetry detection, simplicity, efficiency, rapidity and accuracy. Based on UV Raman spectroscopy, the feasibility of identifying transgenic/non-transgenic soybean oil and other types of edible oil was studied. The UV Raman spectra of five different edible oils (500 samples of each brand of GM/non-GM soybean oil and 100 samples of one kind of rice oil, 2 100 samples in total) in the wavelength range of 3 500~400 cm-1(268~293 nm) were collected. In order to improve the signal-to-noise ratio of spectral data and ensure the accuracy of classification, we used Savitzky-Golay filtering to denoise, adaptive iterative weighted penalty least squares (airPLS) to correct baseline and multiple scattering correction (MSC) to standardize spectrum. According to the UV Raman fingerprint of soybean oil, the main chemical components were analyzed, including fats, proteins and amides. We divided each kind of soybean oil into the training set and test set according to 1∶1, input the training set data into a support vector machine (SVM) for training, and established the best model by 10-fold cross-validation. The recognition accuracy was 99.81%, which had a significant effect on detecting the transgenic soybean. Principal component analysis (PCA) is used for data dimensionality reduction, and 8 principal components were extracted, with a cumulative contribution rate of 74.84%, which can represent most of the characteristics of the original data. On this basis, the preprocessed spectral data were divided into the training set and test set according to 4∶1. The partial least squares regression discriminant analysis (PLS-DA) and 10-fold cross validation method were used to establish the best PLS-DA model of the whole spectrum (the discrimination threshold was set to 0.5) with the accuracy of 70.95%. It is shown that UV Raman spectroscopy can accurately and rapidly identify GM/non-GM soybean oil and rice oil. The study provides an important practical and theoretical basis for the on-site detection of transgenic soybean oil and its food and is of great significance in promoting the development of telemetry identification technology for transgenic products.