Quick Test for Transgenic Components in Rice Using Terahertz Spectra

Xin-Gang Ju,Yuan Zhang,Fei-Yu Lian,Mai-Xia Fu

DOI: https://doi.org/10.1177/0003702818812085

Abstract:The terahertz (THz) spectrum of 0.2-1.6 THz (6.6-52.8 cm-1) was used to identify the existence of transgenic rice Bt63 contents in non-GMO rice using a THz time-domain spectroscopy system. Principal component analysis (PCA) was used to extract the feature data based on the cumulative rate of information contribution ( > 90%); the top four principal components were selected and a radial basis function neural network (RBFNN) method was then trained and used. Three selection radial basis functions including a Gaussian function were used to identify the three types (strong positive, weak positive, and negative). The results show that the samples were identified with an accuracy of nearly 90%; additionally, the positive identification rate was > 87.5% and the negative identification rate reached 100% using the proposed method (PCA-RBF). The proposed approach was then compared with other methods, including back propagation (BP) neural networks and support vector machine (SVM). The results of the comparison show that the accuracy of PCA-RBF method reaches 92% in total and all the rest are < 90% using 100 samples. Obviously, the proposed approach outperforms the other methods and also indicates that the proposed method, in combination with THz spectroscopy, is efficient and practical for transgenic ingredient identification in rice.