An Improved Dual Asymmetric Penalized Least Squares Baseline Correction Method for High-Noise Spectral Data Analysis

Abstract

Baseline drift and noise can blur or even drown out a signal and affect analysis results, especially in multivariate analysis. To address the problem of spectrum denoising and baseline correction, this paper proposes an improved dual asymmetric penalized least squares (IDAPLS) baseline correction method. The proposed method first changes the single parameter λ used for balancing fidelity and roughness in the traditional penalty least squares (PLS) method into a new diagonal matrix Λ and uses the fast convergent inverse tangent S-type penalty function to iteratively estimate the noise level. Then, the diagonal matrix Ψ is introduced into the fidelity of the updated energy spectrum, and the element ψ_i is updated iteratively by using the inverse tangent S-type penalty function. Finally, the baseline of the original signal is obtained when a preset number of iterations or termination criteria are reached. Compared with other methods, IDAPLS solves the problem of underfitted curves when dealing with additive noise that the asymmetric least squares method and adaptive iterative reweighted penalized least squares method would get. The proposed method also retains the advantage of fast PLS and realizes the further approximation of the fitting baseline to the real baseline. Especially, in the case of high noise, this method reduces the error of the traditional PLS method from 30% to less than 5%, which gives a useful reference for nuclear data analysis.

Keywords:

• Baseline correction
• energy spectrum
• penalized least squares

Acknowledgments

The authors are grateful to the team from the Purple Mountain Observatory, Chinese Academy of Sciences, for their valuable suggestions and discussions.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Supported in part by the Fundamental Research Funds for the Central Universities (grant number JZ2019HGTB0082).