The accurate detection of hazardous pesticide residues is crucial for public health. Surface-enhanced Raman scattering (SERS) holds potential but faces practical limitations, including spectral overlap and matrix interference. To address these limitations, we developed a convolutional neural network (CNN)-assisted SERS platform with a hybrid substrate comprising a silver nanostar (AgNS) and hydrophobic silver nanoisland film (AgNIF). This platform synergizes localized surface plasmon resonance with a local concentration effect to achieve high sensitivity, demonstrating a broad linear range and low detection limits for nine pesticides. Coupled with an optimal data preprocessing protocol, our CNN model achieved superior classification accuracy: 99.44% for single pesticides, 98.47% for binary mixtures, 98.09% for ternary mixtures, and 94.60% in spiked tomato samples. Therefore, this work demonstrates a label-free, sensitive and accurate tool for pesticide detection and identification, holding great promise for guiding pesticide application and ensuring food safety.
Di et al. (Fri,) studied this question.