Comparison of Wavelets and Smoothing for Denoising Spectra for Two-Dimensional Correlation Spectroscopy

Abstract

Noise can be severely detrimental to two-dimensional correlation spectroscopy (2D-COS), even at levels as low as 2%. In order to circumvent this problem, spectra must be as free as possible from noise. However, this is not always feasible within the confines of some experiments. Smoothing has been previously used to pretreat the data to reduce noise for 2D-COS. We will show that denoising using wavelets is more beneficial than smoothing and eigenvector reconstruction. The effect of noise is shown on data with 1, 2, and 5% synthetic noise added and then processed with wavelet filtering, smoothing, and eigenvector reconstruction. Lastly, the relative benefits of both smoothing and wavelet filtering on near infrared (NIR) spectra of raw milk are compared.