Enhancing Quantitative Analysis by Laser-Induced Breakdown Spectroscopy (LIBS) with Machine Learning

In the field of laser-induced spectroscopic analysis, leveraging key features within complex spectra is of paramount importance for enhancing the accuracy of quantitative determination. This study enhances laser-induced breakdown spectroscopy (LIBS) accuracy by integrating spectral physics with machine learning. Spectral selection and PCA-LASSO methods extracted characteristic lines satisfying physical-statistical criteria. These features trained CNN and multichannel CNN (MC-CNN) models. MC-CNN achieved superior performance, showing correlation coefficients of 0.9966 (Ni) and 0.9965 (Mn) with residual sums (SSR) of 4.2340 and 0.0521, outperforming single-feature models. Under random noise interference, MC-CNN maintained correlations of 0.9899 (Ni) and 0.9668 (Mn), confirming generalization capability. The fusion of spectral physics and machine learning feature extraction significantly improves LIBS quantification accuracy and model robustness, demonstrating dual advantages in precision and noise resistance.