Integrated surface-enhanced Raman spectroscopy and convolutional neural network for quantitative and qualitative analysis of pesticide residues on pericarp

Pesticide residue poses a significant global public health concern, necessitating improved detection methods. Here, a novel platform was introduced based on surface-enhanced Raman spectroscopy (SERS) to detect ten distinct types of pesticides. Notably, the sensitivity of this approach is exemplified by detecting trace amounts of 50 pM (10 ppt) thiabendazole. The correlation between the characteristic peak intensity of coexisting pesticides and their concentrations displays an exceptional linear relationship (R 2  = 0.9999), underscoring its utility for quantitative mixed pesticide detection. Additionally, qualitative analysis of five mixed pesticides was conducted leveraging distinctive peak labeling. Harnessing machine learning techniques, a model for classifying and predicting pesticides on pericarps was developed. Remarkably, the convolutional neural network achieved classification accuracy of 100 % and prediction accuracy of 99.62 %. This innovative approach accurately identifies and quantifies diverse pesticides, thus offering a feasible scheme for in-situ detection of pesticide residues. Ultimately, this strategy contributes to ensuring food safety and public health.