Multi-way based calibration transfer between two Raman spectrometers.

A standardization algorithm based on the application of Tucker3 models on the tensorized measurement signals is proposed to transfer calibration information between two Raman spectrometers. The secondary instrument in this study is a low cost and portable CCD based unit employing an efficient 532 nm green laser. The primary instrument is a high performance Fourier-transform based laboratory instrument using a low efficiency NIR laser at 1064 nm, albeit with very limited sample fluorescence interference. This work is a first investigation of calibration transfer on Raman spectral data which include different values of fluorescent background from one instrument to the other. The spectra of a small set of calibration samples are measured on both spectrometers. Using the ability of Tucker3 to estimate missing values in tensorized data, we reconstruct the spectrum of a new sample on the primary instrument based on its measured response of the secondary instrument without the need for constructing an explicit transfer model. This way spectra of a prediction sample measured on one spectrometer can be successfully transferred to another spectrometer as if it has been measured directly on the latter. Hence, the task of calibration transfer among instruments is posed as a missing data problem. A discrete wavelet transform is performed to improve the predictive ability. Performance criteria for judging the success of the calibration transfer are reported as the standard error of prediction for estimation of samples in a prediction set. By comparison, the proposed Tucker3 based standardization method shows a better performance as compared to piecewise direct standardization. The method is expected to be applicable for performing calibration transfer using data from instruments other than Raman spectrometers.