A LIBS spectrum baseline correction method based on the non-parametric prior penalized least squares algorithm

• Published in: Analytical methods 2024-07, Vol.16 (26), p.436-4372
• Main Authors: Ma, Shengjie, Xu, Shilong, Chen, Youlong, Dou, Zhenglei, Xia, Yuhao, Ding, Wanying, Dong, Jiajie, Hu, Yihua
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Correlation coefficient; Correlation coefficients; Damage detection; Infrared spectra; Laser induced breakdown spectroscopy; Least squares; Machine learning; Multivariate analysis; Near infrared radiation; Parameters; Quantitative analysis; Raman spectroscopy; Real time; Spectrum analysis
• ISSN: 1759-9660; 1759-9679; 1759-9679
• DOI: 10.1039/d4ay00679h

Laser-induced breakdown spectroscopy (LIBS) has become a popular element analysis technique because of its real-time multi-element detection and non-damage advantages. However, due to factors such as laser-substance interaction and the experimental environment, the measured LIBS spectrum signal contains a continuous background, severely influencing spectrum analysis. In this paper, we propose a LIBS spectrum baseline correction method based on the non-parametric prior penalized least squares (NPPPLS) algorithm. Compared with the traditional Penalized Least Squares (PLS) method, improvements have been made in two aspects. On the one hand, a new weight method with faster convergence is proposed. On the other hand, we combine the Adam algorithm and introduce the RMSE of the baseline correction result at the previous time to constrain the update of the balance parameter, which enables the balance parameter to be adjusted adaptively and no parameter prior is required. The simulation results show that the proposed NPPPLS algorithm can achieve excellent correction results, even with no parametric priors. In addition, the performance of the NPPPLS algorithm is not affected by the initial value of the balance parameter, and the stability and robustness are significantly improved. Finally, we conducted baseline correction of the experimental LIBS spectrum and performed univariate and multivariate analyses. The results show that the quantitative analysis accuracy is improved after baseline correction, and the correlation coefficient R 2 of different elements obtained by the extreme learning machine method of multivariate analysis can reach 0.99, demonstrating a better quantitative analysis result. The simulation and experimental results verify the excellent performance of the proposed NPPPLS algorithm, which can be effectively used to improve the accuracy of quantitative analysis. In addition, this method is also expected to be used for baseline correction of the Raman spectrum, near-infrared spectrum and so on. LIBS spectrum baseline correction method based on non-parametric prior penalized least squares algorithm.