Wavelet analysis applied to removing non-constant, varying spectroscopic background in multivariate calibration

摘要: Multiresolution, the ability to separate signals according frequency, is one of main advantages offered by wavelet transform. However, coarsening resolution associated with this method may be problematic in some applications. The ‘wavelet prism’ (WP) proposed here can split signal into different frequency components, which retain original signal. In conjunction a maximum information gain criterion developed here, new used judge and remove low-frequency non-constant background variation reasonably automatically. paper theory concerning baseline correction are introduced. successfully applied simulated real near-infrared (NIR) spectral data deal for multivariate calibration. Its performance compares favorably current methods removal. appears an efficient removal non-constant, varying spectroscopic background, leading simpler more parsimonious linear model. Copyright © 2002 John Wiley & Sons, Ltd.