为满足食品中反式脂肪酸（TFA）的快检需求, 提出了一种采用近红外漫反射光谱识别含TFA食品的快速无损方法。 采用光纤探头采集完整样品的傅里叶变换近红外漫反射光谱, 应用毛细管气相色谱（GC）法测定食品中TFA的含量作为建模参考数据。 根据食品中TFA含量将食品分为含TFA食品和无TFA食品。 采用偏最小二乘判别（PLSDA）、 支持向量机（SVM）、 簇类独立软模式（SIMCA）和K-最邻近法（KNN）等有监督模式识别方法建立了含TFA食品的识别模型, 并研究了不同光谱预处理方法和建模波段对模型性能的影响。 研究结果表明, PLSDA和SVM两种方法可对含TFA食品进行识别 , 但PLSDA方法识别效果明显优于SVM方法。 其中, 使用与TFA相关波段, 结合标准化和二阶导数预处理所建立的PLSDA识别模型效果最佳, 校正集和验证集识别准确率分别可达96.4%和88%, 具有快速无损识别含TFA食品的可行性。 这种方法不需脂肪提取和研磨等样品预处理, 具有简单、 快速、 无需破坏样品等优点, 非常适合现场或在线快速检测。

A rapid nondestructive method for identifying intact foods containing trans fatty acids (TFA) using diffuse near infrared spectroscopy (NIR) was proposed in the present paper. The diffuse Fourier transform near infrared (FT-NIR) spectra of intact samples were collected by fiber probe, and the reference data of TFA content were determined by Chinese standard method GB/T 22110—2008 (gas chromatography (GC) method). In this work, all the samples were classified into two categories: foods with TFA and foods without TFA according to the TFA content of the foods. The identification models were established by different supervised pattern recognition algorithms including partial least square discriminant analysis (PLSDA), support vector machine (SVM), soft independent modeling of class analogy (SIMCA) and K-nearest neighbor method (KNN) etc. The performances of the established models employing different algorithms, data pretreatments and wavelength bands were compared. The results show that PLSDA and SVM algorithms have the ability of identifying intact foods with TFA, and the performance of identification models established by PLSDA is better than that of SVM. The PLSDA models established by the wavelength bands of 4 138～4 428, 5 507～5 963 and 7 794～8 960 cm-1 which were pretreated with pretreatment methods of auto scaling and second derivative have the best performance. The correct classification percentages of its calibration and validation set are 96.4% and 88%, respectively, which indicates that this method is feasible for the identification of foods with TFA. This NIR method above mentioned has the characteristics of rapidness, non-destruction and easy operation due to the elimination of sample pretreatment such as oil extraction and grinding, therefore it is very suitable for on-line and in-site detection application.