Quantitative analysis of Raman data of pesticide from real sample based on spiking neural networks

A spiking neural network (SNN) based framework was developed for analyzing Raman spectral data. The model was trained to predict the concentration of three pesticides(malachite green, 4-MBA and thiram) by analyzing their surface-enhanced Raman scattering (SERS) spectra at various concentrations, in which the average prediction accuracy could achieve 95%. The performance was superior than convolutional neural network (CNN) and multilayer perceptron (MLP). The the coefficient of determination (R²) value was 0.97 for malachite green detection. Furthermore, this platform was used for detecting thiram from the surface of apple peels with R²at 0.95. The LOD(S/N = 3) of the proposed model for thiram reached 0.37 ppm with recovery value between 96.61%-101.89%. This work demonstrates that the combing SERS with SNN analysis is a feasible, efficient, and precise approach for quantitative detection harmful ingredient from real samples.