Abstract: Design of calibration samples is often used to improve the cost-effectiveness of near-infrared(NIR) spectral analysis. The feasibility of constructing a partial least squares(PLS) model with a parsimonious calibration set for NIR spectral analysis was discussed. Waveband selection was also investigated for improving the predictive ability of the model. The results indicate that, in construction of the quantitative analysis PLS model, the samples in a calibration set should be as similar as possible to the samples in the prediction set. For prediction of the samples with diverse properties, the diversity of the calibration samples is necessary to ensure the predictive ability of the model. On the other hand, waveband selection is proved to be very important for improving the predictive ability of models.

Key words: Near-infrared spectroscopy, Multivariate calibration, Partial least squares, Calibration set design, Wavelength selection