LAUSR

Maize is the main cereal crop in China. In the process of maize planting, the selection of suitable maize varieties is an important link to achieving a high yield. Because the appearance of maize seeds is very similar, it is difficult to accurately identify their species with the naked eye. In order to realize the rapid identification of different varieties of maize seeds, this paper proposes a rapid identification method of maize varieties based on near-infrared (NIR) spectroscopy coupled with locally linear embedding (LLE) and a support vector machine (SVM). The NIR data, preprocessed by multiple scattering correction (MSC), were dimensionally reduced by LLE, a principal component analysis (PCA), and isometric mapping (Isomap), and combined with SVM to establish a maize variety identification model. The results show that the LLE-SVM model has the best performance, whose classification accuracy and kappa coefficient of the test set can reach 100% and 1.00. The classification accuracy and kappa coefficient of the LLE-SVM model are better than the PCA-SVM model and Isomap-SVM model. Therefore, LLE can reduce the complexity of the model and improve the accuracy of the model. It can be used for the rapid identification of maize varieties and provide a new idea for the classification and identification of other agricultural products.