近红外与表面增强拉曼光谱融合技术快速检测花生油中黄曲霉毒素B1

目的 在近红外光谱(near infrared spectroscopy, NIR)与表面增强拉曼光谱(surface-enhanced Raman spectroscopy, SERS)特征层数据融合的基础上构建偏最小二乘回归模型(partial least squares regression, PLSR)实现花生油中黄曲霉毒素B1 (aflatoxin B1, AFB1)含量的快速检测。方法 首先，分别采集待测样本的NIR与SERS光谱。其次，将采集的NIR与SERS光谱分别进行光谱预处理。然后，采用基于希尔伯特-施密特独立准则的变量空间迭代优化算法(Hilbert-Schmidt independence criterion based variable space iterative optimization, HSIC-VSIO)分别筛选NIR与SERS光谱的特征变量。最后，将筛选的特征变量进行融合并构建PLSR模型用于定量检测花生油中AFB1含量。结果 与NIR光谱数据、SERS光谱数据以及NIR与SERS光谱直接融合数据构建的PLSR模型相比，NIR与SERS光谱特征层融合数据构建的PLSR模型具有最佳的预测性能：校正集均方根误差(root mean squared error of calibration set, RMSEC)为0.1569，校正集决定系数(coefficient of determination of calibration set, )为0.9908，预测集均方根误差(root mean squared error of prediction set, RMSEP)为0.1827，预测集决定系数(coefficient of determination of prediction set, )为0.9854，性能偏差比(ratio of performance to deviation, RPD)为8.2761。将本方法与标准方法分别检测真实含有AFB1的花生油样本，结果表明两者的检测性能无显著性差异(P=0.84>0.05)。结论 本方法可实现花生油中AFB1含量的快速、高精度定量检测，也验证了NIR与SERS光谱融合的可行性与有效性。

Rapid determination of aflatoxin B1 in peanut oil by the fusion of near infrared spectroscopy and surface-enhanced Raman spectroscopy

Objective To rapid quantitative analysis of aflatoxin B1 (AFB1) in peanut oil, a partial least squares regression (PLSR) model was built based on the fused characteristic wavelengths which were selected from near infrared spectroscopy (NIR) and surface-enhanced Raman spectroscopy (SERS), respectively. Method First, the NIR and SERS spectra of samples were respectively collected. Second, the collected NIR and SERS spectra were processed by spectral preprocessing techniques. Then, the characteristic wavelengths of NIR and SERS spectra were respectively selected by employing Hilbert-Schmidt independence criterion based variable space iterative optimization (HSIC-VSIO) algorithm. Finally, the selected characteristic wavelengths were fused, and then a PLSR model was built based on the fused characteristic wavelengths for rapid determination of AFB1 in peanut oil. Results Compared with the PLSR models built by NIR spectra, SERS spectra, and the data directly fused by NIR and SERS spectra, the PLSR model built based on the fused characteristic wavelengths which were respectively selected from NIR and SERS spectra possessed superior prediction performance: the value of root mean squared error of calibration set (RMSEC) is 0.1569, the value of coefficient of determination of calibration set ( ) is 0.9908, the value of root mean squared error of prediction set (RMSEP) is 0.1827, the value of coefficient of determination of prediction set ( ) is 0.9854, and the value of ratio of performance to deviation (RPD) is 8.2761. The proposed method and the standard method were employed to determine the content of AFB1 in real peanut oil samples, respectively. The results showed that there was no significant difference between these two methods (P=0.84>0.05). Conclusion The proposed method is a promising alternative for rapid and high-precision determination of the content of AFB1 in peanut oil. The feasibility and effectiveness of the fusion of NIR and SERS spectra were also verified.