Corn protein classification and detection method based on near infrared spectral features combined with TCN model

As a climate-smart crop, high-quality genetic improvement of corn plays an important strategic role in ensuring global food supply. Protein is a key indicator for evaluating corn quality. Therefore, accurate detection of corn protein content is of great significance for directional regulation of corn quality and smart cultivation decision-making. In view of the problems existing in the current corn protein detection research, such as damaged samples, low precision, and complicated procedures. This paper proposes a corn protein detection model based on near-infrared (NIR) spectroscopy combined with temporal convolutional networks. Firstly, Savitzky-Golay (SG) was applied to preprocess the data to effectively remove the spectral scattering information. Then, a Genetic Algorithm (GA) was used to extract eight effective characteristic wavenumbers from the 1845 preprocessed wavenumbers. Finally, the multivariate time analysis characteristics of the time series model Temporal Convolutional Network (TCN) were used to construct a corn protein detection model with an accuracy of 95.35 %. Compared with Back Propagation neural network (BP), Support Vector Machine (SVM), Convolutional Neural Networks (CNN), Transform, and CNN-transform, the accuracy of this model was improved by 25.58 %, 21.45 %, 15.35 %, 41.86 %, and 39.54 %, respectively. This method provides a new idea and approach for the detection of corn protein and other crop proteins. • The proposed SG-GA-TCN corn protein detection model has high accuracy. • The proposed TCN based detection model performs better than BP, SVM, CNN, Transform, and CNN-transform. • This paper provides a new idea and method for corn protein detection. • The research results provide technical support for the rapid detection of crop proteins.