Evaluation of optimal wavelet decomposition level for quantitative calibration of near infrared spectra

In order to avoid selecting decomposition level for quantitative calibration of near infrared spectra in discrete wavelet transform empirically or semi-empirically, a scheme which can be selected optimal decomposition level was proposed. In this scheme, the raw spectral was decomposed by wavelet transform, the approximation coefficients of each decomposition level as input set for partial least square (PLS) to built calibration model, root mean square error of validation (RMSEV) was employed as objective function to find the optimal decomposition level using the validation set, and the approximation coefficients which obtained lowest RMSEV is the optimal decomposition level. The proposed method was demonstrated in determination of moisture in wheat by near infrared spectroscopy.