Stochastic regression modeling of chemical spectra

• A novel approach to separate signal from estimated noise
• An operator-independent replacement for smoothing and filtering
• Use of the Epanechnikov kernel key to generating the proper regression

A stochastic regression model is presented that separates signal from noise in chemical spectra. Spectra are decomposed into additive contributions from signal and from estimated noise. Numerical results on sample spectra are presented and suggest that this strategy offers an effective and computationally efficient framework for comprehensive noise estimation and analysis. From this analysis more effective methods of feature extraction in chemical spectra can be created.