Performance improvement of temperature compensation in near infrared analysis of orange sweetness by applying direct standardization

Abstract

Near infrared spectroscopy is a non-destructive technique used for measuring and analyzing chemical compositions in an organic sample. The calibration equation and spectrum are used for calculating the prediction result. In this case, the spectrum provides very important data; therefore, the accuracy of the near infrared prediction system depends on the sample preparation because the spectrum is sensitive to physical property conditions such as sample temperature. When the sample temperature has changed, the absorption peak will be shifted nonlinearly in both the absorption value and wavelengths around 840 nm and 940 nm (in the short regions). Consequently, if applying a calibration model developed from spectra of a constant sample temperature by using a linear multivariate data analysis to predict the samples with different temperature conditions, the average of difference between actual values and predicted values (bias) will occur. Therefore, the objective of this research was to develop a spectra temperature compensation method namely the temperature compensation coefficient method by applying direct standardization algorithm. By the use of temperature compensation coefficient, the temperature effect can be solved and the accurate prediction results can be obtained. Moreover, the performance of temperature compensation coefficient was investigated and compared with the fixed temperature and three compensation methods, such as generalized least squares weighting, external parameter orthogonalization, and global calibration. The results indicated that temperature compensation coefficient method and the global calibration gave the best result with high accuracy of the lowest bias at 95% confident level.