食用油酸值与过氧化值近红外光谱模型转移研究

在使用近红外光谱技术进行食用油酸值与过氧化值检测时,仪器制造与检测环境的差异导致不同仪器建立的校正模型无法共享。为解决食用油酸值与过氧化值模型转移问题,使用125个食用油样本于主机建立偏最小二乘校正模型,采用光谱空间转换法进行模型转移,并与斜率/截距算法、直接标准化算法、分段直接标准化算法、极限学习机自编码器算法进行对比。结果表明,采用光谱空间转换法进行模型转移后,验证集酸值与过氧化值的预测均方根误差分别从0.583 6 mg/g和15.801 0 mmol/kg降低到了0.167 0 mg/g与9.989 3 mmol/kg,说明光谱空间转换法可以有效应用于食用油酸值与过氧化值间的模型转移,使不同仪器之间实现模型共享,这对于近红外光谱应用于食用油品质快速检测具有实际意义。

Transfer of Near-infrared Spectroscopy Model of Edible Oil Acid Value and Peroxidation Value

Near-infrared spectroscopy (NIRS) detection technology has been widely used in the field of edible oil quality detection due to its features of simple operation, no damage, fast detection and analysis. However, due to differences in manufacturing and testing environment, instrument aging, replacement of accessories and other factors, the spectral response of the same sample on different instruments was not completely consistent, resulting in the failure to share calibration models established by different instruments. To solve the problem of edible oil quality detection of acid value and peroxide value model transfer problem, transfer model was established by using the spectral space transformation method, which realized the calibration transfer of acid value and peroxide value between different instruments, and the slope/bias algorithm, direct standardized algorithm and piecewise direct standardization, extreme learning machine auto-encoder algorithm were compared. The VERTEX-70 Fourier near infrared spectrometer and MATRIX-F on-line process analysis Fourier spectrometer were used. Totally 125 edible oil samples were used in the experiment, VERTEX-70 was used as master instrument and MATRIX-F as slave instrument to establish the multivariate calibration model of acid value and peroxide value. The results showed that the prediction root-mean-square error of the validation set of acid value and peroxide value was reduced from 0.5836mg/g and 15.8010mmol/kg to 0.1670mg/g and 9.9893mmol/kg by using spectral space transformation method for model transfer. Obviously, spectral space transformation can be effectively applied to model transfer between edible oleic acid value and peroxidation value, so as to realize model sharing between different instruments. The problem of time and energy consumption caused by re-modeling can be avoided, which was of great significance for the application of NIRS in the rapid detection of edible oil quality.