Fast quantitative detection of thiram using surface-enhanced Raman scattering and support vector machine regression

Abstract

As a novel and ultrasensitive detection technology that had advantages of fingerprint effect, high speed and low cost, surface-enhanced Raman scattering (SERS) was used to develop the regression models for the fast quantitative detection of thiram by support vector machine regression (SVR) in the paper. Meanwhile, three parameter optimization methods, which were grid search (GS), genetic algorithm (GA) and particle swarm optimization (PSO), were employed to optimize the internal parameters of SVR. Furthermore, the influence of the spectral number, spectral wavenumber range and principal component analysis (PCA) on the quantitative detection was also discussed. Firstly, the experiments demonstrate the proposed method can realize the fast and quantitative detection of thiram, and the best result is obtained by GS-SVR with the spectra of the range of characteristic peak which are processed by PCA. And the effect of GS, GA, PSO on the parameter optimization is similar, but the analysis time has a great difference in which GS is the fastest. Considering the analysis accuracy and time simultaneously, the spectral number of samples over each concentration should be set to 50. Then, developing the quantitative model with the spectra of range of characteristic peak can reduce analysis time on the promise of ensuring the detection accuracy. Additionally, PCA can further reduce the detection error through reserving the main information of the spectra data and eliminating the noise.