Quantitative analysis of near infrared spectra by wavelet coefficient regression using a genetic algorithm

Application of FTIR Spectroscopy to Agricultural Soils Analysis

“…Although wavelet decomposition by itself does not produce a compressed representation of the original data, data reduction can be achieved by eliminating the wavelet coefficients that do not contain valuable information. Various approaches have been reported in the literature for selecting the most relevant coefficients, such as eliminating all "small" coefficients using for instance either thresholding (Kai-man Leung et al 1998;Ehrentreich 2002), entropy (Kai-man Leung et al 1998), mutual information (Alsberg et al 1998), maximum likelihood (Leger&Wentzell 2004), or genetic algorithms (Depczynski et al 1999), or retaining only the coefficients with the highest variance (Trygg&Wold 1998) as depicted in Figure 6. Once data compression has been achieved, the remaining coefficients can be used as input variables for a neural network that creates a non-linear mapping between these inputs and the property (or properties) of interest.…”

Genetic algorithms in chemometrics and chemistry: a review

“…Depczynski et al [73] devised a method for multicomponent analysis by near-infrared spectrometry by combining wavelet coefficient regression with a GA.…”

Multivariate calibration of analytical signals by WILMA (wavelet interface to linear modelling analysis)

“…The discrete wavelet transform (DWT) has been applied as a preprocessing tool in multivariate calibration of near-IR spectra [14][15][16][17], mid-IR spectra [18], Raman spectra [19], fluorescence data [20], X-ray powder diffraction spectra [21], process variables [22] and electroanalytical signals [23].…”

“…Where DWT is used for feature selection, two approaches are most often proposed: (1) the wavelet coefficients are thresholded by using criteria based on the evaluation of PLS weights [18] or of PLS regression coefficients [15]; (2) the wavelet coefficients are previously ranked by their variance [16,20,21] or by their squared correlation coefficient with the response variable [17]; then the subset giving the most stable or the best performing regression model is selected.…”