The global increase in plastic waste, exceeding 400 million tons annually, underscores the urgent need for efficient plastic sorting and recycling. Laser-induced breakdown spectroscopy (LIBS) shows potential in this area, but its practical application is limited by challenges such as plasma fluctuations and low robustness. To address these limitations, we introduce laser-induced plasma acoustic (LIPA) signals and propose the frame-segmentation LIPA (FLIPA) algorithm to enhance LIBS analysis. This innovative algorithm reduces the number of variables in LIPA by 99% while optimizing computational efficiency and classification accuracy. Additionally, a multimodal fusion technique, LIBS-FLIPA, is developed to integrate LIBS and FLIPA at the feature level. The results indicate that LIBS-FLIPA significantly improves classification accuracy, robustness, and generalization, effectively mitigating overfitting risks. This study provides novel, to the best of our knowledge, solutions to challenges in LIBS analysis and proposes an innovative approach for robust plastic sorting, advancing the methodologies of LIBS research.
To address the critical challenges of flow field heterogeneity and stochastic signal fluctuations in the online monitoring of industrial aerosols, this paper proposes a novel, to the best of our knowledge, Laser-Induced Breakdown Spectroscopy (LIBS) enhancement technique based on acoustic-optical multimodal synergistic perception. Departing from conventional monitoring paradigms that rely solely on spectral data, this study introduces the acoustic signatures of laser-induced shockwaves as a novel physical dimension to elucidate the transient interaction mechanisms between the laser and microscopic particles. The developed Physics-Aware Adaptive Gated Fusion (P-AGF) network enables deep mining and dynamic complementarity of spectral and acoustic information. Experimental results demonstrate that the P-AGF model can precisely identify the microscopic interaction states between the laser and particles in complex aerosol scenarios, such as soldering smoke, achieving a test set accuracy of 92.22%. This method overcomes unimodal detection limits in mixed-phase media, significantly enhancing LIBS stability and robustness for intelligent industrial monitoring.