1.College of Optics & Electronics Engineering, University of Shanghai for Sci and Tech, Shanghai 200093, China;
2.Laboratory of NIR- Applications, Zhejiang University of Sci and Tech, Hangzhou, Zhejiang 310023, China;
3. National Quality Supervision and Testing Center for Rice Wine, Shaoxing, Zhejiang 310027, China
Abstract：To improve the fitting degree between structural characteristics of near-infrared spectroscopy and non-linear quantitative regression model, a quantitative calibration method by the wavelet projection pursuit is proposed. The method reduces the noise of NIR wavelet coefficients by gaussian mixture model noise estimation method. After finding optimal low-dimensional projection of the spectrum wavelet coefficients, the calibration model was fitted by the polynomial ridge regression. The experiment sets up wine alcohol prediction model of near infrared spectroscopy by wavelet projection pursuit regression method, the R2 and RMSECV are 0.957 and 0.37838, comparing with quantitative effect of multiple linear regression and partial least squares regression, spectral wavelet projection pursuit method has better prediction performance that can be more effectively applied quantitative analysis of near infrared spectroscopy.
2013, Vol. 34 
