An Intelligent Photocatalytic Self-Cleaning SERS Platform with Machine Learning for On-Site Detection of Multiple Antibiotics

The widespread use of antibiotics in aquaculture has led to persistent residues in aquatic environments, necessitating the development of sensitive, sustainable, and on-site detection methods. This study presented an intelligent surface-enhanced Raman spectroscopy (SERS) platform that integrated photocatalytic self-cleaning and machine learning-assisted quantification. The platform was constructed on a hierarchical composite substrate (OCC@TiO 2 @AgNPs), which synergistically combined electromagnetic enhancement from AgNPs and chemical enhancement from the TiO 2 layer. The platform exhibited excellent stability (RSD = 7.8% over 28 days), outstanding batch-to-batch reproducibility (RSD = 4.2%), and effective self-cleaning capability via TiO 2 photocatalysis, enabling over 95% signal recovery across five consecutive detection−regeneration cycles. Using an optimized machine learning model, the platform achieved ultra-low detection limits of 1.47 pM for enrofloxacin (ENR), 0.52 pM for sulfadiazine (SD) and 2.68 pM for midecamycin (MID), with a wide linear response range of 10−10 7 pM. Validation with real environmental samples (lake, river and shrimp) confirmed high accuracy, with recovery rates of 92.49−107.10% and precision (RSD below 8.76%), demonstrating strong resistance to matrix interference. This SERS platform provided a practical and sustainable solution for on-site monitoring of multiple antibiotic residues in aquatic systems, with significant potential for applications in environmental surveillance and food safety.