High-precision identification transgenic sugarcane using active terahertz low-frequency excitation

This study proposes a novel, environmentally friendly, accurate, rapid, and cost-effective technique for the classification and identification of transgenic sugarcane, combining terahertz (THz) time-domain spectroscopy with an Isomap-Deep Convolutional Neural Network (Isomap-DCNN). Different types of transgenic sugarcane derived from the same receptor material exhibit similar THz absorbance spectra, posing significant challenges for accurate categorization. The Tukey window function was first applied to the THz absorbance spectra of five transgenic sugarcane varieties, and the processed data were subsequently used for feature extraction with t-distributed Stochastic Neighbor Embedding (t-SNE), LargeVis (LV), and Isomap. To enhance the robustness and generalization capability of the model, the Kennard-Stone (KS) algorithm was employed to partition the training and testing sets. The processed key "fingerprint" features were then input into a Decision Tree (DT) classifier and a Deep Convolutional Neural Network (DCNN) classifier, respectively. Notably, the Isomap-DCNN model achieved 100 % accuracy in identifying the five transgenic sugarcane varieties. The Isomap-DCNN model, constructed based on THz spectroscopy, is expected to provide a fast, high-precision, environmentally friendly, and cost-effective technique for the classification and identification of transgenic sugarcane. This approach will aid in the development and implementation of regulatory standards, enhance the safety and efficacy of transgenic sugarcane, and promote public transparency regarding information about transgenic sugarcane. • Terahertz time-domain spectroscopy for the detection of different types of transgenic sugarcane. • The KS algorithm partitions datasets to enhance model generalization and robustness. • Identification of transgenic sugarcane events using the Isomap-Deep Convolutional Neural Network classification model.