Multimodal integration of dual-wavelength fluorescence and elemental emission features for event-resolved classification of aerosolized samples

To mitigate spectral overlap and amplitude variability in ultraviolet laser-induced fluorescence (LIF) measurements of bioaerosols, we propose an interpretable mid-level data fusion framework for event-resolved aerosol classification. Within each aerosol-triggered event, dual-wavelength LIF spectra (263 and 355 nm) and laser-induced breakdown spectroscopy (LIBS) signals are acquired sequentially to achieve event-level cross-modal alignment, capturing complementary fluorescence responses as molecular fingerprints and elemental emissions of Mg, Ca, Na, and K as orthogonal compositional descriptors. To ensure statistically valid performance estimates under event-to-event variability and heterogeneous feature dimensionality across modalities, all spectral preprocessing, dimensionality reduction, and feature-level fusion steps are performed under strict training–test separation and evaluated using 10-fold cross-validation. Based on an event-aligned dataset of ten fluorescent aerosol surrogates (5000 spectra), the multimodal fusion significantly improves the discrimination of spectrally-similar but chemically-distinct categories. For instance, the accuracy for distinguishing structural polysaccharides (chitin) from protein samples increases from about 46% using LIF alone to about 87% after fusion. The overall classification accuracy reaches 86.7% while maintaining millisecond-level inference efficiency. These results demonstrate that the synergy between molecular fluorescence and atomic stoichiometry provides a useful analytical basis for trigger-defined event-level aerosol classification under the present laboratory conditions, and offers a methodological basis for future validation in more complex environments.