Quantitative analysis of acetamiprid and thiacloprid in green tea using enhanced SERS and neural networks

Neonicotinoids (NEOs) are widely used in agricultural cultivation, and the presence of residual NEOs poses a serious threat to consumer health. Therefore, trace detection of NEOs is particularly important. In this study, a response surface experimental design was employed to develop AgNP with high enhancement factor and surface height anisotropy, effectively enhancing the Raman signals of acetamiprid (ACE) and thiacloprid (THI) through densely populated hotspots on the surface. Gaussian simulation calculations were used to assign peaks in the ACE and THI spectra. Subsequently, convolutional neural network, backpropagation neural network, and AlexNet neural network were employed to identify the ACE, THI, and mixed component spectra, successfully distinguishing the spectra of the three compounds and achieving qualitative analysis of unknown spectra. Furthermore, genetic algorithm-partial least squares was used to model the spectral data, achieving a correlation coefficient of 0.97. The method was applied for the detection of spiked tea samples, achieving a precision RSD of 4.84 %. Therefore, this sensor combined with intelligent algorithms can be used for spectral identification and trace detection of mixed pesticide samples with structurally similar compounds. •The response surface experiment was used to assist the design of Ag SERS substrates.•Peak assignments of Acetamiprid and Thiacloprid were obtained by Gaussian simulation.•Three types of neural networks were used for SERS spectral identification.•GA-PLS was successfully applied in the quantitation of multi-component.•The method can be applied to the analysis of other mixed components in real samples.