Skip to Main Content
This paper presents the preliminary research in detecting pesticide residues with the newly developed terahertz technique. Experiments were carried out on cyfluthrin n-hexane solutions in the concentration range of 1-20 ?g/ml with terahertz time-domain spectroscopy (THz-TDS). We employed four calibration methods for the regression between the absorbance and the concentration of the solutions, i.e., linear regression methods of simple linear regression (SLR), partial least squares (PLS), and nonlinear regression methods of simple least-squares support vector machine (LS-SVM), and the combination of PLS and LS-SVM (PLS-LS-SVM), which utilizes the features extracted by PLS from the original absorbance spectra as the input of LS-SVM. To analyze the antinoise abilities of these models, we collected two data sets of the same series of samples on two different days, i.e., one with higher signal-to-noise ratio (SNR) and the other with lower SNR. Separate models were built on them. We find that the PLS and PLS-LS-SVM models present much better prediction and antinoise abilities than the SLR and simple LS-SVM models, whereas PLS-LS-SVM outperforms PLS, showing much less dependence on the selection of the number of the features extracted. To further analyze the influence of the selection of the frequency range, we built different PLS-LS-SVM models by varying the frequency ranges between 0.5 and 1.5 THz, and find that the model performance dramatically varies with the selection of different frequency ranges, which may be due to the unevenly distributed SNR and the spectroscopic interaction variations in different frequency ranges. The selection of the whole frequency range of 0.5-1.5 THz results in relatively low prediction error with either the higher or the lower SNR spectra and will thus be selected for further modeling. The overall results indicate that the THz-TDS plus PLS-LS-SVM method proves to be an effective tool for the quantitative analysis of cyfluthrin n-hexane sol- - utions and may provide a new technique for further pesticide residue detection.