Qualitative and Quantitative Analyses of Potassium Sorbate in Milk Powder using Terahertz Spectra

Using the terahertz time-domain spectroscopy system (THz-TDS), the potassium sorbate content in milk powder has been investigated. Firstly, the THz time-domain spectra of the samples were transformed into frequency spectra by Fourier transform, and then, through a set of simplified formulas of the optical parameters, the THz absorption spectra and refractive index spectra were obtained. Based on the characteristic absorption peak of potassium sorbate, a quantitative analysis model of potassium sorbate in milk powder was constructed using ordinary least squares (OLS) linear regression. In the case of low content (<1%) of potassium sorbate, a special THz absorption band (0.887-1.000 THz) was selected to construct a partial least squares (PLS) model that is called Sub-PLS. In addition, for qualitative analysis, we used partial least squares discrimination analysis (PLS-DA). Experimental results show that the proposed Sub-PLS model is superior to the PLS model based on full spectrum in accuracy, and the proposed PLS-DA is superior to other nonlinear regression methods such as support vector machine and neural network.