Correcting Multivariate Calibration Model for near Infrared Spectral Analysis without Using Standard Samples

Abstract

Model transfer is essential for practical applications of near infrared (NIR) spectroscopy because differences may exist between the spectra measured using different instruments. For correcting the calibration models in cases where standard samples are unavailable, a method is proposed based on the relationship between the prediction error and the spectral difference of the instruments. In this method, a partial least squares (PLS) model, named the primary model, was constructed using the calibration spectra measured with a primary instrument, and the model was then used to predict the spectra from a secondary instrument. Error may arise in the prediction and the predicted error is obviously caused by the difference between the spectra of the two instruments. Assuming that there is a linear relationship between the spectral difference and the prediction error, a correction PLS model can be built to predict the error using the spectral difference. The predicted error obtained with the correction model can be used to correct the error of the primary model. Two NIR datasets, of pharmaceutical tablets and tobacco leaf samples, measured with two instruments were used to test the performance of the method. The results show that, for both datasets, the prediction error for the spectra of the secondary instrument of the primary model can be satisfactorily corrected.