Multi-information fusion of characteristic spectral lines for enhanced quantitative accuracy in laser-induced breakdown spectroscopy

Abstract

When employing Laser-Induced Breakdown Spectroscopy (LIBS) for the quantitative analysis of heavy metal elements in soil, conventional quantitative analysis methods typically rely solely on single spectral line intensity information. The quantitative results are prone to fluctuations in excitation source energy, environmental parameter variations, and self-absorption effects, resulting in relatively low detection accuracy. To overcome these limitations and improve the accuracy of quantitative analysis, this study first introduced a decision filter based on the Median Absolute Deviation (MAD) algorithm for data processing, thereby improving data reliability. Subsequently, by analyzing trends in peak intensity, Spectral Peak Area (SPA), and full width at half maximum (FWHM) of spectral lines as a function of elemental concentrations, and evaluating the linear fitting performance of these three distinct types of information, the relative advantages of various spectral line information in quantitative analysis were assessed based on excitation principles. Building on this, a nonlinear Support Vector Regression (SVR) model was employed to integrate intensity, SPA, and FWHM from characteristic spectral lines. This integration facilitated information complementarity, thereby improving the accuracy of quantitative analysis. Experimental results demonstrated that the nonlinear regression model, incorporating multiple spectral line information, significantly enhanced quantitative analysis accuracy compared with linear regression models utilizing only single types of spectral line information. Specifically, the root mean square error (RMSE) for chromium (Cr) and lead (Pb) was 0.036 wt% and 0.026 wt%, respectively, with an average relative error (ARE) of 8.624% and 5.733%. The fitting coefficients (R²) exceeded 0.98 for both elements. By integrating the multi-dimensional information of characteristic spectral lines and employing a nonlinear SVR model for calibration, this study effectively overcomes the limitations inherent in traditional quantitative analysis relying on single-peak information. This approach significantly enhances the accuracy and robustness of LIBS technology for quantitative detection of heavy metals in soil, providing an effective solution for high-accuracy, multi-information-fused LIBS quantitative analysis in complex environments.