This paper investigates the influence of wavelet family, length and number of resolution levels on the performance of multivariate calibration models obtained in the wavelet domain. Twenty-one physical and chemical properties of diesel, gasoline, corn and wheat were determined by near/mid infrared spectrometry employing partial least-squares (PLS) and stepwise regression (SR) in the original and wavelet domains. Through proper selection of the wavelet transform settings, average RMSEP reductions of 8.2% (PLS) and 27.0% (SR) were obtained with respect to the original domain. However, the SR models presented considerable sensitivity with respect to the choice of transform settings. In this case, an analysis of variance indicated that the number of resolution levels is the most important factor to be considered.
multivariate calibration; wavelet transform; analysis of variance; mid and near infrared spectrometry; food and fuel analysis
