Identification of official rhubarb samples based on THz spectra and least squares support vector machine

Terahertz time-domain spectroscopy (THz-TDS) combined with emphatic orthogonal signal correction (EOSC) and least squares support vector machine (LS-SVM) has been applied to establishing qualitative analysis model for identifying 41 official and unofficial rhubarb samples. The correction factors of the model have been determined by bootstrapped Latin-partitions. The spectra were pretreated by autoscaling and Savitzky - Golay smoothing (the window size is 5, the polynomial order is 3), and a comparison of two pretreatment methods for the classified accuracy was implemented. The results showed that the identification accuracy of 97.84±1.62% could be accomplished by using the pretreatment methods of autoscaling and EOSC, which is higher than the classified accuracy of 87.45 ± 3.03% by using Savitzky - Golay smoothing method. The proposed method was proved to be a convenient, non-polluting, accurate, and nondestructive approach for identifying rhubarb samples. The developed procedure can be easily implemented for controlling quality in rhubarb production.