Fig. 1. Network structure of GAN

Fig. 2. Network model of DNN

Fig. 3. THz spectral curves of riboflavin before and after S-G filtering. (a) Before filtering; (b) after filtering

Fig. 4. Comparison before and after data preprocessing. (a)?(c) Before preprocessing; (d)?(f) after preprocessing

Fig. 5. THz spectra of 10 substances

Fig. 6. Data graphs generated by GAN under different iterations. (a) Original spectral data graph; (b) data graph generated by 1000 iterations; (c) data graph generated by 5000 iterations; (d) data graph generated by 10000 iterations; (e) data graph generated by 100000 iterations; (f) data graph generated by 200000 iterations

Table 1. Comparison of sample size between original and extended data sets

Table 2. Recognition accuracy rate of different classifiers