过氧化值的快速、 准确检测对食用油脂的品质及其食品安全控制具有重要意义。 近红外光谱技术是一种理想的过氧化值测量手段, 但校正模型的建立需要耗费大量的资源。 旨在通过近红外光谱信息与油脂过氧化物间的关系分析, 探索对不同种类、 不同等级植物油建立同一校正模型的可行性, 以不同等级的大豆油和菜籽油为研究对象, 结合二维相关光谱技术对两种植物油的近红外光谱进行分析, 通过间隔偏最小二乘法选择过氧化值通用模型的最佳检测波段, 考察了正交信号校正(OSC)、 标准正态变量变换(SNV)和二阶导数(SD)对两种植物油过氧化值校正模型的影响, 比较了主成分回归(PCR)、 偏最小二乘法(PLS)和支持向量机回归(SVR)三种建模方法的预测效果, 构建了大豆油(一级+三级)、 菜籽油(一级+三级+四级)、 一级油(大豆油+菜籽油)、 三级油(大豆油+菜籽油)四种通用模型。 结果显示: (1)近红外光谱能够检测植物油过氧化值的变化情况, 对应的光谱信息主要分布于1 700~2 200 nm区域; (2)通用模型最佳的波段、 预处理方法和建模方法分别为1 700~2 200 nm、 SD法和PLS法; (3)四种通用模型中一级植物油(大豆油和菜籽油)的过氧化值通用模型具有较好预测结果, 其预测均方根误差(RMSEP)、 决定系数(R2)分别为0.412和0.920, 与一级的大豆油和菜籽油单一模型相比, 预测精度相差不大。 研究表明生产工艺过程相差不大的一级植物油间有可能建立准确性高的通用模型。 此外, 为了扩展通用模型的性能, 需要不断用新产品对模型进行及时更新。

The rapid and accurate detection of peroxide value is of great significance to the quality of edible oils and food safety control. Near-infrared (NIR) spectroscopy is an ideal method for measuring peroxide values, but the establishment of calibration models requires a lot of resources. In this paper, based on the relationship between near-infrared spectroscopy and lipid peroxide value, the feasibility of establishing a general calibration model for measuring peroxide value in different types and levels of vegetable oils was studied. First, different grades of soybean oil and rapeseed oil were studied. The near-infrared spectra of the two vegetable oils were analyzed by two-dimensional correlation spectroscopy, and the optimal detection band of the general model for peroxide value was selected by interval least squares method (iPLS). Then, the effects of orthogonal signal correction (OSC), standard normal variable transformation (SNV) and second derivative (SD) on the prediction precision of general models were investigated. Further, the prediction performances of three modeling methods, including principal component regression (PCR), partial least squares (PLS) and support vector machine regression (SVR), were compared in detail. At last, four general prediction models for soybean oil (including first-grade and third-grade), rapeseed oil (including first-grade, third-grade and fourth-grade), first-grade vegetable oils (including soybean oil and rapeseed oil) and third-grade vegetable oil (including soybean oil and rapeseed oil) were constructed. The experimental results showed that the change of peroxide value of vegetable oil could be detected by NIR spectroscopy technology, and its spectral changes mainly resided in the region from 1 700 to 2 200 nm. The optimal band, preprocessing method and modeling method of the general model were 1 700~2 200 nm, SD method and PLS method, respectively. Among the four general models, the one for the first-grade vegetable oils (including soybean oil and rapeseed oil) can get better performance. The root means square error of prediction (RMSEP) and the square correlation coefficient (R2) are 0.412 and 0.920, respectively. Compared with the models of first-grade soybean oil or rapeseed oil, the prediction results are roughly the same. It meant that it was feasible to establish a general model with high accuracy between vegetable oils with similar production processes for reducing the workload of repetitive modeling. In addition, in order to expand the versatility of the general model, it was necessary to continuously update the model with new kinds of vegetable oils.