Classification of wheat grain varieties using terahertz spectroscopy and convolutional neural network

Wheat quality and quantity differ in diverse climates. Therefore, it is essential to identify the variety before purchasing and warehousing. In this study, a study on variety discrimination for 12 wheat varieties (stronggluten wheat, medium -gluten wheat, weak -gluten wheat) using the Terahertz time -domain spectroscopy (THzTDS) technology in combination with a Convolutional neural network (CNN). Firstly, the original Time -domain spectra (TDS) of wheat in the range of 0.1-2.0 THz were acquired, and the Frequency domain spectra (FDS), the absorption coefficient spectra in the range of 0.2-1.0 THz were obtained through Fourier Transform. Then, Competitive adaptive reweighted sampling (CARS) algorithms were applied to screen the feature spectrum. Finally, the Support vector machine (SVM), the Least square support vector machine (LS-SVM), the Backpropagation neural networks (BPNN) and CNN models were constructed using feature spectral data. By comparing the four models, it was found that the calibration set accuracy and prediction set accuracy of the CNN model reached 98.7% and 97.8% respectively, with an error recognition rate of only 2.2%. The research results show that combining THz-TDS technology with CNN has the advantages of accurate recognition and high efficiency. It can identify different wheat varieties and can be used for seed classification and quality detection.