Field‑deployable transformer‑enhanced SERS platform for rapid screening of multi‑pesticide residues with real‑sample validation

Pesticide contamination in real agricultural environments often involves mixed residues, generating complex spectral interference that challenges conventional Raman-based detection. Rapid, accurate, and field-deployable analytical tools capable of identifying multi-component pesticide mixtures across diverse fruit matrices are therefore urgently required. Here, we present an integrated surface-enhanced Raman scattering–Transformer platform for on-site multi-pesticide screening. Reproducible gold nanoparticles provide a highly sensitive and stable surface-enhanced Raman scattering substrate, and a comprehensive dataset of 5979 spectra—covering single, binary, ternary, and quaternary mixtures in apple, peach, pear, and grape extracts—captures realistic variability. The Transformer achieves superior performance, with an accuracy of 98.56% on the standard test set and 97.16% on real fruit extracts, including matrix-specific accuracies of 98.18% for apple, 97.89% for pear, 96.59% for peach, and 96.57% for grape; notably, the precision, recall, and F1-score on the standard test set all exceeded 0.985, five-fold cross-validation yielded 97.74% ± 1.02%, confirming the robustness and reproducibility of the method. Attention-weight visualization offers qualitative interpretability by highlighting relevant spectral regions. On-site orchard trials in Tianjin, China, further validate the robustness and practical applicability of the workflow. Overall, this study advances surface-enhanced Raman scattering-based pesticide detection by establishing one of the largest multi-pesticide spectral datasets to date, demonstrating strong cross-matrix generalization, and delivering an integrated portable Raman–Transformer pipeline suitable for rapid field screening and routine regulatory surveillance.