基于近红外光谱技术的赣南茶油掺假快速鉴别Rapid identification of Gannan oil-tea camellia seed oil adulteration based on near infrared spectroscopy

为了探索基于近红外光谱技术快速无损鉴别掺假油茶籽油的可行性，以赣南茶油为研究对象，通过掺入不同植物油如玉米油、花生油、菜籽油、葵花籽油和大豆油等制备掺假油茶籽油，应用近红外光谱技术采集其光谱特征信息，对比不同预处理方法和主成分数，并结合线性和非线性建模方法建立油茶籽油掺假鉴别模型，以识别准确率(纯油茶籽油样品和掺假油茶籽油样品被正确判别的比例）、灵敏度（纯油茶籽油样品被正确判别为纯油茶籽油的比例）、特异性（掺假油茶籽油样品被正确判别为掺假油茶籽油的比例）作为模型的评价指标，优选出最佳模型。结果表明：二阶微分联合线性判别分析(SD-LDA）模型为最优线性模型，标准正态变量变换联合人工神经网络(SNV-ANN）模型为最优非线性模型，两个模型的识别准确率、灵敏度、特异性分别为 97.58%、100%、97.33%和98.99%、100%、98.88%。SNV-ANN 模型鉴别效果优于SD-LDA模型，说明非线性模型更适于油茶籽油掺假判别，该模型能更准确地鉴别油茶籽油是否掺假。 In order to explore the feasibility of rapid and non-destructive identification of adulterated oil-tea camellia seed oil based on near infrared spectroscopy, Gannan oil-tea camellia seed oil was selected as the research object, and the adulterated oil-tea camellia seed oil was prepared by blending different vegetable oils such as corn oil, peanut oil, rapeseed oil, sunflower seed oil and soybean oil. The spectral characteristics of the adulterated oil-tea camellia seed oil were collected by near infrared spectroscopy, and different pretreatment methods and main components were compared and determined. The identification model of oil-tea camellia seed oil adulteration was established by combining linear and nonlinear modeling methods. The identification accuracy(the percentage of pure and aduterated oil-tea camellia seed oil samples correctly identified, sensitivity (the percentage of pure oil-tea camellia seed oil samples correctly identified as pure oil-tea camellia seed oil) and specificity (the percentage of adulterated oil-tea camellia seed oil samples correctly identified as adulterated camellia seed oil) were used as the evaluation indexes of the model to select the best model. The results showed that the second derivative-linear discriminant analysis (SD-LDA) model was the optimal linear model, and the standard normal variable transformation-artificial neural network (SNV-ANN) model was the optimal nonlinear model, their identification accuracy, sensitivity and specificity were 97.58%, 100%, 97.33% and 98.99%, 100%, 98.88% respectively. The identification effect of SNV-ANN model was better than that of SD-LDA, which indicated that the nonlinear model was more suitable for the identification of oil-tea camellia seed oil adulteration, and the model could more accurately identify whether the oil-tea camellia seed oil was adulterated.