Application of wavelet transforms to improve prediction precision of near infrared spectra

Near-infrared spectroscopy (NIRS), with the characteristics of high speed, non-destructiveness, long-range detection, high precision and reliable detection data, etc., is a pollution-free, rapid, quantitative and qualitative analysis method. Whether or not NIRS can be used in products analysis, depends on its prediction precision. In this paper, a method based on wavelet transform, which is used to analyze near-infrared spectra, is discussed with the purpose of improving precision of reducing sugar prediction in vinegar. The results show that the feature spectra of original signals can be separated by wavelet transform with variant scales. By the comparison of the multiple regression analysis for the absorbed spectrum, the following conclusions can be made: the least uncertainty of prediction difference is realized when the original spectra are analyzed by wavelet transform with a scale of 7. In this case, the uncertainty of prediction difference is decreased 0.058 compared with the original spectra. The dimensions of the regressed independent variables are also important to prediction precision. To avoid over-fitting in calibration, the selected dimensions of independent variables should not be excessive. By the research, this method can really improve the prediction precision by properly selecting transform scales and regression dimensions.