Atmospheric carbon concentration scenario classification based on the fusion of spectral and acoustic modalities
ShiHao Liu, Yu Zhang, Jun Feng, Wenhan Gao, TianLong Li, Yuzhu Liu
Abstract
Accurate carbon concentration monitoring is vital for climate change mitigation and environmental management. While laser-induced breakdown spectroscopy (LIBS) offers rapid, non-destructive detection, it is affected by plasma fluctuations and environmental interference. Laser-induced plasma acoustic signals (LIPA) can capture CO 2 molecular characteristics, and in this study, we propose a multimodal LIBS-WLIPA method that fuses spectral and acoustic data for stable classification of four gas scenarios. A wavelet-based WLIPA algorithm was developed to efficiently process noisy acoustic signals, reducing variables by 99% while preserving key information. Using LIBS-WLIPA, we compared six machine learning models and assessed the contributions of LIBS and WLIPA features. Results show that LIBS-WLIPA markedly improves detection accuracy, robustness, and generalization, with Logistic Regression, Random Forest, XGBoost, and CatBoost achieving 97.5% accuracy. This method offers an efficient solution for gas environment classification and expands the application potential of LIPA technology in environmental monitoring.