Abstract: Terahertz time domain spectra for 40 flour samples with different contents of benzoyl peroxide were measured, and frequency domain spectra, refractive spectra and absorption spectra were obtained in the range of 0.2–2.0 THz. The time domain spectra indicated that the terahertz wave amplitude value of the flour samples with benzoyl peroxide was reduced as compared to reference signal, and the frequency domain signal was delayed to different degrees. The frequency domain spectra indicated larger amplitude and lower intensity of terahertz wave absorption for flour samples with higher content of benzoyl peroxide, demonstrating that benzoyl peroxide could reduce terahertz wave absorption in flour. The refractive spectra indicated that refractive index decreased gradually with the increase in frequency, but it increased with increasing content of benzoyl peroxide in flour. The absorption spectra indicated that benzoyl peroxide in flour samples had a significant characteristic absorption peak at 1.6 THz. Therefore, terahertz spectrum can be used for the identification of benzoyl peroxide in flour. Partial least squares (PLS), support vector regression (SVR) and least squares support vector regression (LS-SVR) were used separately to establish a calibration model by using the absorption spectral data of 40 flour samples as input variables. The determination coefficients (R2) of both the PLS and SVR models were lower than 0.9, with relative standard deviations (RSD) greater than 10%, while the R2 of the LS-SVR model was 0.986, with RSD of 4.30%, indicating that terahertz spectroscopy can be used to detect the content of benzoyl peroxide in flour.

Key words: terahertz spectroscopy, benzoyl peroxide, qualitative, least squares support vector regression, quantitative, rapid nondestructive detection