MoS2@Ag nanoflower-based SERS substrate coupled with transformer network for sensitive detection and intelligent classification of fluoroquinolone antibiotics in milk.

Abstract

Fluoroquinolone antibiotics (FQs) residues in dairy products pose a persistent challenge to food safety. This work establishes an integrated analytical platform combining a three-dimensional molybdenum disulfide‑silver nanoflower (MoS2@Ag NFs) SERS substrate with a transformer network based deep learning model for the sensitive detection and intelligent classification of FQs in milk. The prepared MoS2@Ag NFs substrate with a 1T/2H mixed-phase MoS2 and a hierarchical structure densely decorated with silver nanoparticles (Ag NPs), exhibits a high enhancement factor and achieves nanomolar-level detection limits for four representative FQs. To overcome the discrimination challenge posed by the highly similar spectral signatures of different FQs, a transformer network is employed to automatically extract decisive spectral features via its self-attention mechanism, which results in an exceptional classification accuracy exceeding 99.82%. The integrated platform enables sensitive and intelligent monitoring of FQs in milk, providing a promising strategy for on-site detection of antibiotic residues in complex food matrices.