Baseline Correction Based on a Search Algorithm from Artificial Intelligence.

Area of Science:

• Spectroscopy
• Analytical Chemistry
• Computational Science

Background:

• Polynomial-like baselines are common in spectral data.
• Iterative polynomial fitting is a popular but flawed baseline correction method, especially with strong peaks.
• Existing methods struggle with baseline errors caused by strong peaks at spectrum endpoints or within the data.

Purpose of the Study:

• To develop a novel search algorithm-based (SA) baseline correction method.
• To address limitations of iterative polynomial fitting in spectral baseline correction.
• To improve accuracy in estimating baselines for spectra with strong peaks.

Main Methods:

• Spectra are smoothed using a moving average to reduce noise.
• Smoothed spectra are divided into sections based on Chebyshev nodes.
• A search algorithm minimizes mean absolute error to remove peak data points, forming the corrected baseline.

Main Results:

• The SA method demonstrated lower baseline errors compared to Lieber and Mahadevan-Jansen, adaptive iteratively reweighted penalized least squares, and improved asymmetric least squares methods.
• Experiments used simulated and real Fourier transform infrared (FTIR) and Raman spectra.
• SA effectively corrects baselines even in the presence of strong spectral peaks.

Conclusions:

• The proposed search algorithm-based method offers a more accurate approach to spectral baseline correction.
• SA effectively mitigates errors associated with strong peaks in polynomial-like spectra.
• This method provides a robust alternative for spectral data processing in analytical chemistry.