Classification of transgenic corn varieties using terahertz spectroscopy and convolutional neural network

Corn is the largest grain crop in China. With the increasing use of transgenic corn in agricultural production, accurately identifying transgenic corn varieties has become crucial for ensuring food safety and quality. This study combines terahertz time-domain spectroscopy (THz-TDS) technology with a convolutional neural network (CNN) to classify five transgenic corn varieties. First, the time-domain spectra of transgenic corn were collected, and the frequency-domain spectra and absorption coefficient spectra were calculated. Next, abnormal data were removed using the isolation forest algorithm. The data were then preprocessed using the first derivative (FD) and principal component analysis (PCA) to extract effective spectral features. Based on these features, classification models were constructed using support vector machine (PCA-SVM), grid search optimized SVM (PCA-GS-SVM), particle swarm optimization SVM (PCA-PSO-SVM), and hippopotamus optimization SVM (PCA-HO-SVM). Additionally, a one-dimensional CNN (1D-CNN) model was built using the full-band spectral data. The results indicate that the performance of the 1D-CNN model with FD preprocessing was significantly improved, with an accuracy of 96.68 % and a misclassification rate of only 3.32 %. The study effectively classifies transgenic corn with high accuracy, showing potential for widespread application in quality detection.