PRISM: a peak region identification and signature mapping method based on laser-induced spectroscopy-acoustics fusion for metal corrosion discrimination

To overcome the single-modal limitations of Laser-Induced Breakdown Spectroscopy (LIBS) for identifying diverse metal corrosion types in industrial settings, we introduce a multi-modal approach by incorporating acoustic signals, termed Laser-Induced Spectroscopy-Acoustics. We propose a novel dimensionality reduction method, the Peak Region Identification and Signature Mapping Algorithm (PRISM), tailored to extract features from the characteristic peaks of both the optical spectra and acoustic signals. A Random Forest (RF) model trained with PRISM-extracted features achieved a classification accuracy of 96.67%, outperforming four other machine learning methods. Feature contributions were evaluated using SHapley Additive exPlanations (SHAP). Our work demonstrates that this multimodal fusion significantly enhances classification performance and has great potential for industrial applications.