Automatic baseline correction by wavelet transform for quantitative open-path Fourier transform infrared spectroscopy.

A technique for automatically correcting the baseline of open-path Fourier transform infrared spectra is described in which the spectra are decomposed into high-frequency (details) and low-frequency (discrete approximations) information using a wavelet transform. After an appropriate number of iterations, n, the discrete approximation simulates the baseline of the spectrum. By setting the nth approximation to zero and reversing this process, the reconstructed signal contains only the high-frequency components in the original signal on a baseline that is approximately flat and at zero absorbance. When small molecules such as NH3 and CH4 are quantified by partial least-squares (PLS) regression, this process decreases the number of eigenvectors required for the analysis by two or three, increasing the ruggedness of the prediction. The baseline that is calculated is approximately the same for spectra measured at resolutions of 1, 2, 4, 8, and 16 cm(-1). Surprisingly, the baseline that is calculated is not totally smooth, with artifacts that are caused by the analyte. The amplitude of the artifacts is directly proportional to the concentration of the analyte, so the accuracy of PLS regression is not degraded.