20 September 2002 Artificial Neural Network and application in calibration transfer of AOTF-based NIR spectrometer
Author Affiliations +
Chemometrics is widely applied to develop models for quantitative prediction of unknown samples in Near-infrared (NIR) spectroscopy. However, calibrated models generally fail when new instruments are introduced or replacement of the instrument parts occurs. Therefore, calibration transfer becomes necessary to avoid the costly, time-consuming recalibration of models. Piecewise Direct Standardization (PDS) has been proven to be a reference method for standardization. In this paper, Artificial Neural Network (ANN) is employed as an alternative to transfer spectra between instruments. Two Acousto-optic Tunable Filter NIR spectrometers are employed in the experiment. Spectra of glucose solution are collected on the spectrometers through transflectance mode. A Back propagation Network with two layers is employed to simulate the function between instruments piecewisely. Standardization subset is selected by Kennard and Stone (K-S) algorithm in the first two score space of Principal Component Analysis (PCA) of spectra matrix. In current experiment, it is noted that obvious nonlinearity exists between instruments and attempts are made to correct such nonlinear effect. Prediction results before and after successful calibration transfer are compared. Successful transfer can be achieved by adapting window size and training parameters. Final results reveal that ANN is effective in correcting the nonlinear instrumental difference and a only 1.5~2 times larger prediction error is expected after successful transfer.
© (2002) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Wenbo Wang, Chengzhi Jiang, Kexin Xu, Bin Wang, "Artificial Neural Network and application in calibration transfer of AOTF-based NIR spectrometer", Proc. SPIE 4927, Optical Design and Testing, (20 September 2002); doi: 10.1117/12.471415; https://doi.org/10.1117/12.471415

Back to Top