Detection of pesticide residues by sensor arrays fused from SERS spectra of various substrates combined with deep learning

In this study, a rapid, accurate and intelligent method for detecting pesticide residues was developed by sensor arrays fused from surface-enhanced Raman spectroscopy (SERS) spectra of various substrates combined with deep learning. First, spectral data of pesticide residues were collected using conventional substrates with various morphologies. Then, the sensor array is acquired using different fusion methods on the spectral data, specifically concatenation and parallel methods. Finally, the intelligent and precise detection of chlormequat chloride and paraquat is implemented in a regression model constructed through the sensor array combined with machine learning and deep learning. In this experiment, the lowest detection limits for chlormequat chloride and paraquat detected are 0.03 mg/L and 0.003 mg/L, respectively. The regression model which was constructed using the sensor arrays obtained from the concatenation fusion method combined with the MobileNet obtained the best prediction results, with the coefficients of determination of prediction of 0.9865 and 0.9846 for chlormequat chloride and paraquat, and root-mean-square errors of prediction of 0.2254 and 0.0174, respectively. Overall, the method we proposed achieves rapid, intelligent and accurate detection of chlormequat chloride and paraquat, with the advantages of low cost and simple operation, which has great application potential in trace detection of pesticide residues.