A comparison of noise models in a hybrid reference spectrum and principal components analysis algorithm for Raman spectroscopy

摘要: Raman spectroscopy exploits the scattering effect to analyze chemical compounds with use of laser light. spectra are most commonly analyzed using ordinary least squares (LS) method. However, LS is known be sensitive variability in analyte and background materials. In a previous paper, we addressed this problem by proposing novel algorithm that models expected variations as well signals. The method was called hybrid principal component analysis (HLP) used an unweighted Gaussian distribution model noise measured spectra. show fact follows Poisson improve our accordingly. We also approximate weighted model, which enables more efficient solver algorithm. To reflect generalization from hereon call reference spectrum components (HRP) compare performance HRP (HRP-G), (HRP-P), (HRP-WG) models. Our experiments both simulated data experimental acquired serial dilution Raman-enhanced gold-silica nanoparticles placed on excised pig colon. When only signal zero-mean random (as examined data), HRP-P consistently outperformed HRP-G HRP-WG, latter coming close second. Note scenario, were equivalent. presence mean spectra, three algorithms significantly LS, but performed similarly among themselves. This indicates that, significant modeling such important than optimizing model. It suggests for real data, HRP-WG provides desirable trade-off between accuracy computational speed. Copyright © 2013 John Wiley & Sons, Ltd.