An Entropy-Based Method of Background and Noise Removal for Analysis of Near-Infrared Spectra

A new algorithm (WFCE) was proposed for simultaneously eliminating background and noise based on wavelet packet transform (WPT) and information entropy theory. At first, WPT algorithm and reconstruction algorithm were employed to split the raw spectra into different frequency components. Then the information entropy of each frequency component was calculated, showing the uncertainty to the measured analyte concentration. At last, based on comparison of information entropy, the importance of each frequency component to the whole spectra was evaluated and the suitable wavelet components representing background and noise can be determined for removal. WFCE algorithm was validated by measuring the original extract concentration of beer using the NIR spectra. The results show that the prediction ability and robustness of models obtained in subsequent partial least squares calibration using WFCE were superior to those obtained using other algorithm, and the root mean square errors of prediction can decrease by up to 38.6%, indicating that WFCE is an effective method for elimination of background and noise.