Identification of Starch Types by Using Terahertz Time-Domain Spectroscopy Combined with SG-KNN

The terahertz (THz) spectra of four different types of starch were investigated by terahertz time domain spectroscopy (THz-TDS). Subsequently, standard normal variate (SNV) and savitzky-golay (SG) were used to remove the interference of spectral noise, and then a K-Nearest Neighbor (KNN) method was applied to identify the four types of starch. The results showed that the samples were identified with average accuracy, Kappa, and F1 scores for the test set were $0.9375,0.9158$, and 0.9383, respectively. This method was also contrasted with logistic regression (LR). The comparisons demonstrated that SG-KNN surpassed the other methods and indicated that the integration of THz technology with SG-KNN was an effective approach for identifying starch types.