This study proposed a rapid, non-destructive method for identifying Wuchang Daohuaxiang rice based on near-infrared (NIR) spectroscopy combined with machine learning algorithms. NIR spectra of different varieties of rice were collected. First-order derivative was determined as the best spectral preprocessing method using partial least squares regression (PLSR). Two dimensionality reduction methods, principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE), were compared, and five machine learning models including artificial neural network (ANN), K-nearest neighbors (KNN), random forest (RF), decision tree (DT), and Naive Bayes (NB) were constructed for variety classification and comparison. The results showed that t-SNE improved the Calinski-Harabasz index by 1078.0051, demonstrating better clustering performance. After t-SNE dimensionality reduction, the performance of all five models was superior to that without dimensionality reduction. The average classification accuracy was 95.78%. The accuracy of the NB model was improved most effectively (by 18.89%). The random forest model showed the best classification performance, with prediction accuracy and precision of 98.89% and 98.96%, respectively. This method provides a rapid, non-destructive solution for identifying Wuchang Daohuaxiang rice, which will contribute to brand protection and safeguarding consumer rights, and also offers a new approach for the identification of other geographical indication agricultural products.

Food spoilage caused by mold has posed a major threat to grain quality and national food safety. Mold can also produce mycotoxins damage to the nutritional value of cereals, and even be harmful to human health. It is highly required for efficient and reliable detection. Traditional detection of the mold in grains can often involve time-consuming laboratory tests that rely heavily on specialized equipment. It is very necessary to realize the rapid on-site assessments. Alternatively, near-infrared spectroscopy (NIRS) can be expected to rapidly detect the mold of agricultural products, due mainly to the detection speed, non-destructive testing, repeatability, and easy online analysis. The purpose of this study was to establish a mold detection model in rice using NIRS. A systematic investigation was also implemented to rapidly distinguish rice from the different levels of mold. A dataset of 960 samples was taken from four varieties with different mold degrees (2018 Mudanjiang 27, 2019 Mudanjiang 27, Longjing long grain aromatic rice, and Muxiang 1). Qualitative discrimination models were then realized for the different degrees of mold contamination. The first derivative was combined with 9-point smoothing and factorization during preprocessing, in order to obtain the qualitative discriminant model with high accuracy. The mean S value greater than 1 represented the excellent performance to distinguish between mildewed and non-mildewed rice. The accuracy of the model was further verified by a leave-one cross-validation. The accuracy reached 93%. In addition, 300 independent data sets of rice samples were also utilized with the different degrees of mildew. The total number of mold colonies was quantitatively characterized using NIRs. A discrimination model was then established by vector normalization and partial least squares (PLS) method. Some indexes were calculated to evaluate the accuracy of cross-validation root-mean-square error (RMSECV), determination coefficient (R²), and performance deviation ratio (RPD). The prediction root-mean-square error (RMSEP) was also used to evaluate the accuracy of the model. The results showed that the values of RMSECV, R², RPD, and RMSEP were 0.470, 0.904 5, 3.24, and 0.45, respectively. The first three parameters indicated the high precision of the model, while the latter indicated the high accuracy of the model. Therefore, the improved model with high accuracy was achieved to rapidly predict the mold in grains. The mold was the main influencing factor on the variation of the NIR spectrum after optimization. On the contrary, there were relatively small effects of rice variety and harvest year on the spectral characteristics. Therefore, the NIRs can be highlighted to detect mold contamination in rice, regardless of variety or time differences. This finding can provide a strong reference for the rapid prediction of the mildew degree or amount of rice in different transportation using NIR spectroscopy. Online real-time monitoring equipment can also be offered to monitor the grain mildew in containers.