基于中红外光谱法检测煎炸油极性组分

目的 利用中红外光谱技术实现对煎炸油极性组分的快速检测。方法 根据SPXY法对煎炸油中红外光谱数据进行样本划分,从而得到校正集和预测集。采用SG+一阶导数预处理手段,利用竞争自适应重加权算法(competitive adaptive reweighted sampling, CARS)进行特征提取,建立煎炸油极性组分含量的偏最小二乘回归(partial least squares regression, PLSR)预测模型,并利用误差反向传播算法(error back proragation, BP)对模型进行优化。结果 BP神经网络法建立的模型校正集决定系数(coefficient of determination, R_²)为0.8073,校正集均方根误差(root mean square error of calibration,RMSEC)为0.0325,预测集R_²为0.7665,预测集均方根误差(root mean square error of prediction, RMSEP)为0.0443。结果表明,经BP神经网络算法优化后,均方根误差明显减小,提高了预测模型的准确性。结论 结合BP神经网络算法的中红外光谱技术是一种检测煎炸油极性组分的有效方法,为食用油油品品质的快速检测提供理论指导和技术支撑。

Detection of polar components in frying oil based on mid-infrared spectroscopy

Objective In this paper, mid-infrared spectroscopy has been proposed to realize polar components detection in special frying oil. Methods According to SPXY method, mid-infrared spectrum data of frying oil samples was divided into two parts, which included calibration set and prediction set. Moreover, competitive adaptive reweighted sampling (CARS) algorithm was used for feature extraction after SG+1st derivative preprocessing. The prediction model of polar components had been established by partial least squares regression (PLSR) methods, which was optimized by error back proragation(BP). Results Coefficient of determination(R2) and root mean square error of calibration (RMSEC) based on BP neural network model were 0.8073 and 0.0325, R2 and root mean square error of prediction (RMSEP) were 0.7665 and 0.0443, respectively. Results suggested that the root mean square error decreased after the optimization of BP neural network algorithm, which resulted in the improvement of model prediction accuracy. Conclusion Combined with BP neural network algorithm, mid-infrared spectroscopy technology provides a useful method to detect polar components in special frying oil, which supplied theoretical guidance and technical support for rapid quality detection of edible oil.