Full-spectrum LIBS quantitative analysis based on heterogeneous ensemble learning model

Abstract

Laser-induced breakdown spectroscopy (LIBS) technology is widely used in fields such as analytical chemistry, materials science, and environmental monitoring. Modeling the quantitative relationship between component contents and spectral data is a key step in LIBS analysis. However, traditional regression methods commonly use individual regression model, which are difficult to comprehensively and reasonably utilize the information in the spectra, resulting in limitations in full-spectrum multicomponent regression. This paper proposes a heterogeneous ensemble learning (HEL) model, selecting four heterogeneous sub-models: CNN, Lasso, Boosting, and FNN, for full-spectrum LIBS quantitative regression analysis. HEL can fully leverage the strengths of different models by using Bayesian weighting strategy, thereby improving the performance of LIBS quantitative analysis. Experimental results show that the proposed HEL regression model has better accuracy and stability compared to the existing models.