|Sohn, Mi Ryeong|
|Barton Ii, Franklin|
|De Haseth, James|
Submitted to: Applied Spectroscopy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/26/2009
Publication Date: 10/1/2009
Citation: Sohn, M., Himmelsbach, D.S., Barton Ii, F.E., De Haseth, J.A. 2009. Transfer of Calibration for barley quality from dispersive intrument to fourier transform near-infrared instrument. Applied Spectroscopy. 63 (10). p. 1190-1196. 2009. Interpretive Summary: In our previous studies, we developed near-infrared calibration models for quality assessment of barley as a source of fuel ethanol production. Calibration models developed on one instrument need to be transferred to other instruments for more practical use of the models. However, when an existing model is applied to spectra measured on different instruments, problems can occur. The best way to solve the problem is to measure all samples in each instrument and to develop a new model for each instrument. This procedure is, however, very time- and labor-consuming when data set is huge. If samples were collected over several years, it is possible that they could be no longer available to use due to changing of the constituents or the lack of existence of the samples. Hence, the ability to transfer calibrations from one instrument to another has importance in many ways, saving a significant amount of work and cost as well as having access to a large data set for which to compare new samples. The goal of this study was to transfer a near-infrared calibration model for barley quality developed on one instrument to another instrument having different resolution and response, using advanced chemometric techniques. This paper describes how the calibration transfer was successfully accomplished by choosing a proper transfer sample set and by updating the transferred model. The research results will have interest for a broad audience working in the field of algorithm development for spectroscopic data and of its transfer between instruments.
Technical Abstract: This study deals with transferring the near-infrared (NIR) calibration models for quality assessment of barley between two instruments with different resolutions and number of data points, a Fourier transform instrument (master) and a dispersive instrument (slave). A file of spectra from 206 ground barley samples was tested and the constituents of primary interest were starch and protein. The spectral data from the slave were transferred to the master and calibrated. The models were applied to the prediction set on master and the transferred results were evaluated on a basis of root mean square error of prediction (RMSEP). For the standardization between the instruments, piecewise direct standardization method was used and two sets of transfer samples, a subset of barley samples and 30 sealed powder standards were used, respectively. The transferred models resulted in a standard error of cross-validation of 1.52% and 0.25% for starch and protein, respectively. For the standardization using the barley subset, the models resulted in an RMSEP of 1.59% and 0.28% for starch and protein, respectively, indicating a successful calibration transfer with an acceptable accuracy. For the standardization using the sealed powder standards, the models resulted in an RMSEP of 3.40% and 0.61% for starch and protein, respectively, indicating the transfer was not as successful with large biases. However, with the addition of few spectra from the master instrument to the the transferred data, new models could be generated to predict barley samples measured on master instrument with a good accuracy, giving an RMSEP of 1.49% and 0.24% for starch and protein, respectively.