Hexagonal Au Nanostructure SERS Metasurface for AI-Driven Detection of Pesticide Residues in Real Food Samples

Pesticide residues in food remain a major threat to human health and ecosystems, yet routine monitoring still relies on centralized, multistep analytical workflows which are poorly suited to rapid and field-deployable detection. In this work, we introduce a rationally designed hexagonal honeycomb metal-insulator-metal (MIM) plasmonic metasurface which functions as a robust, wafer-scale surface/plasmon-enhanced Raman spectroscopy (SERS) platform for pesticide quantification in real food matrices. The MIM honeycomb architecture simultaneously creates highly concentrated electromagnetic hotspots at the excitation wavelength and a plasmonic antenna effect that radiates the Stokes-shifted Raman signals back, effectively multiplying the Raman signal and enabling sensitive detection of multiple fungicides and insecticides directly in cucumber extracts. We show that characteristic vibrational fingerprints can be reliably captured for several representative pesticides (metalaxyl, boscalid, famoxadone, thiamethoxam, etoxazole, cypermethrin) across realistic concentration ranges and in the presence of complex matrix backgrounds, achieving subppm limits of detection that approach or fall below current regulatory maximum residue limits. To convert raw spectra into actionable readouts, we integrate our process flow with a deep feed-forward (DFF) artificial intelligence model pipeline that performs automated spectral preprocessing and supervised learning for both pesticide identification and residue-level classification with respect to regulatory thresholds. This AI-enabled MIM-SERS platform establishes a generalizable route toward compact, high-throughput instruments for multiresidue pesticide surveillance in real food samples, with broader implications for molecular diagnostics and environmental monitoring.