Trace-level detection of free polycyclic aromatic hydrocarbons based on magnetic driving and deep learning-assisted recognition.

Polycyclic aromatic hydrocarbons (PAHs) are persistent organic pollutants with strong carcinogenicity and bioaccumulation, posing serious threats to aquatic ecosystems and human health. However, the sensitive and accurate detection of trace-level PAHs in water remains a significant challenge. In this study, a surface-enhanced Raman spectroscopy (SERS) strategy integrating magnetic-driven enrichment, cyclodextrin-specific molecular capture, and deep learning-based spectral analysis was developed for the rapid detection of PAHs in water. A ring-shaped Fe₃O₄@Au substrate (Mag_SERS) modified with thiolated β-cyclodextrin (SD) enabled host-guest inclusion and promoted favorable molecular orientation under applied magnetic field. Electromagnetic field simulations using COMSOL and DFT calculations revealed the presence of abundant electromagnetic "hot spots" and identified an Au → SD → PAH charge-transfer pathways, both contributing to substantial Raman signals enhancement. To address severe spectral overlap among structurally similar PAHs, a Sparrow Search Algorithm-optimized CNN-LSTM-Attention (SCLA) model was constructed, achieving a classification accuracy exceeding 98% and a detection limit as low as 10-8 M. The proposed method provides a rapid, intelligent, and field-deployable SERS platform for the accurate monitoring of PAHs in complex environmental water samples.