A novel system for precise identification and explainability analysis based on multimodal learning combining laser-induced breakdown spectroscopy and laser-induced plasma acoustic signals.

Abstract

This study presents an innovative approach to identify copper types using Laser-Induced Breakdown Spectroscopy (LIBS) in conjunction with Laser-Induced Plasma Acoustic (LIPA) signals. Traditionally overlooked, plasma acoustic signals can indeed provide valuable insights into plasma characteristics essential for copper identification. This study pioneers a cross-modal learning technique, integrating LIBS and LIPA signals, and employs a Support Vector Machine (SVM) for classification. To enhance feature extraction, Principal Component Analysis (PCA) reduces data dimensionality, while SHapley Additive exPlanations (SHAP) assess feature contributions, aiding feature selection. The combined model demonstrates high identification accuracy, and the interpretability analysis deepens our understanding of feature roles in copper detection. This framework not only boosts LIBS-based identification accuracy but also advances the theoretical foundation for multi-modal data fusion in material analysis.