Near-infrared characteristic spectrum variable selection method based on window competitive self-adaptive reweighted sampling strategy

Abstract

The invention discloses a near-infrared characteristic spectrum variable selection method based on a window competitive self-adaptive reweighted sampling strategy. The method comprises the steps of establishing an mXn matrix through near-infrared spectrum data X collected by m samples under n near-infrared wavelengths, wherein each column of the matrix represents spectrum variables of the m samples under one wavelength; equally dividing the n spectrum variables into N windows in the wavelength direction; building a PLS model by randomly selecting the spectrum variables of the mX80% samples andcorresponding sample target component data y, obtaining regression coefficients, performing statistics on the regression coefficients in each window, and using a mean of absolute values of the regression coefficients as a basis index of whether the window is reserved or not; deleting the spectrum variables of the windows with the minimum regression coefficient absolute value means sequentially, and until the number of the reserved windows is defined in the specification, stopping the process; building the PLS model by using the residual spectrum variables and the sample target component datay, and calculating a root mean square error of cross validation (RMSECV); and building a final model by taking the spectrum variable corresponding to the minimum value of the RMSECV as a near-infraredcharacteristic spectrum variable, and performing analysis.

Description

technical field [0001] The present invention relates to non-destructive analysis in the field of analytical chemistry, in particular to a window competitive adaptive reweighted sampling strategy method (window competitive adaptive reweighted sampling, WCARS) for near-infrared characteristic spectrum variable selection. Background technique [0002] With the rapid development of near-infrared instruments and chemometric methods, near-infrared spectroscopy has been widely used. Near-infrared spectroscopy reflects the frequency multiplication and combination frequency information of material absorption, and contains rich information on the composition and molecular structure of most types of organic matter. Compared with traditional methods, near-infrared spectroscopy has the advantages of strong penetrating power, no need for complicated pretreatment operations, no damage to samples, and long-distance online detection through optical fibers. Therefore, it is widely used in man...

AI Technical Summary

Problems solved by technology

However, this method is prone to inconsistent results between the calibration set and the prediction set during the calculation process.

This is because the CARS algorithm relies too much on the cross-validation results of the calibration set and does not consider the synergy between adjacent variables.

Images