Feature-enhanced dual-input transformer for LIBS quantitative analysis of minor-content elements

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Materials Science

Background:

• Laser-Induced Breakdown Spectroscopy (LIBS) offers rapid, in situ elemental analysis.
• Accurate quantification of minor elements in LIBS is challenging due to weak signals, noise, and spectral interferences.
• Current methods struggle with feature extraction, relying on incomplete physics-based features or data-driven models that may overfit or suppress trace signals.

Purpose of the Study:

• To develop a novel method for robust quantitative analysis of minor-content elements using LIBS.
• To overcome limitations of existing approaches in handling weak spectral signatures and matrix effects.
• To improve the accuracy and reliability of elemental quantification in complex materials.

Main Methods:

• A feature-enhanced dual-input transformer model was developed.
• A cross-attention mechanism aligned physical atomic spectra with LIBS spectra to guide feature selection.
• Self-attention processed LIBS spectra to learn informative wavelength contributions, and a gated attention fusion balanced physics-guided and data-driven insights.

Main Results:

• The model significantly improved quantitative analysis of minor elements in LIBS.
• Key performance indicators like R-squared, RMSE, and MAE showed substantial enhancements compared to existing methods.
• The approach effectively integrated physical priors and data-driven learning to enhance target element features.

Conclusions:

• The feature-enhanced dual-input transformer provides a robust strategy for accurate minor-element quantification in LIBS.
• This method effectively combines physics-guided and data-driven approaches for improved spectral analysis.
• The study enables more reliable elemental measurements of complex materials across diverse applications.