A Machine Learning‐Decoded Plasmonic Nanofinger Array for Discriminative SERS Sensing of Nitroaromatic Explosives

ABSTRACT Combining machine learning with surface‐enhanced Raman scattering (SERS) offers a powerful paradigm for pattern‐recognition‐based biochemical sensing applications. However, the discriminative detection of nitroaromatic explosives remains a formidable challenge due to their weak affinity for plasmonic nanostructures, inherently low Raman cross‐sections, and severe spectral overlap among structural analogues. Here, a machine learning‐decoded plasmonic nanofinger array is proposed for nanomolar‐level SERS discrimination of nitroaromatic explosives. Capillary forces drive self‐approaching nanofingers to confine analytes at saddle points. Templated by highly‐ordered colloidal nanospheres, the substrate achieves high reproducibility (RSD < 5.2%), suppressing signal noise from morphological heterogeneity. A cross‐reactive sensor array functionalized with different thiolated aromatic reporters is developed to amplify the spectral differences induced by various nitroaromatic explosives. The cross‐reactive SERS spectra are concatenated into a “superprofile,” providing a comprehensive fingerprint for each analyte. Crucially, machine learning chemometric models are employed to decode these complex, high‐dimensional spectral datasets into diagnostic fingerprints, enabling unambiguous identification and discrimination of nitroaromatic explosives at nanomolar concentrations. The platform achieves 100% discrimination accuracy for four major nitroaromatic compounds, demonstrating exceptional specificity. This synergistic combination of engineered plasmonic substrates and intelligent data analytics significantly advances SERS toward sensitive and specific trace detection of nitroaromatic explosives.