Intelligent identification of ethyl paraoxon and methyl parathion based on surface enhanced Raman spectroscopy

Surface-enhanced Raman scatters (SERS) spectroscopy is a novel detection technology which has advantages of fingerprint, high sensitivity, simple pretreatment and strong anti-interference for water and has been widely used for the analysis of organophosphorus pesticide residues. Furthermore, the intelligent species identification and quantitative analysis of organophosphorus pesticides can be achieved by combing with chemometrics methods. In the actual detection, the classification accuracy of conventional algorithms are limited for the recognition of SERS spectra of some structural analogues. The paper introduces a novel algorithm by the fusion of boosting and support vector machine (SVM) to improve the recognition accuracy of similar SERS spectroscopy of pesticides (ethyl paraoxon and methyl parathion). In the paper, the spectra of the above two pesticides from 600 to1800 cm^-1 were firstly measured using dynamic SERS, and the baseline drift of spectra was deducted through adaptive penalty least-square method. The high frequency burr was reduced by the polynomial smoothing. Finally, the classification model was respectively constructed using SVM and AdaBoost-SVM which combined the discrete AdaBoost (the one implementation of boosting) with SVM, and the algorithm performance was quantitatively evaluated using the 5-fold interaction validation method with the classification accuracy. The experimental results show that the overall classification identification of Adaboost-SVM is significantly superior to SVM, and the accuracy increases by nearly 4.23%. Additionally, during the tuning of Co and gstep for the Adaboost-SVM, the effect on the classification performance is relatively small. The phenomenon demonstrates Adaboost-SVM has the excellent robustness.