Coupling Machine Learning with Clusterization-Triggered Emission for Geographical Origin Tracing of Rice

Tracing the geographical origin of rice is of great significance in protecting the rights and interests of consumers and legitimate producers, as well as ensuring food safety. Here, we propose the combination of machine learning (ML) and clustering-triggered emission (CTE) of rice for identifying the geographical origin of rice. Upon studying the CTE-related emission properties of rice samples of different origins, it was found that minor differences in the intrinsic emission properties of rice samples did exist but were difficult to be directly used for geographical origin tracing. Machine learning was thus introduced to explore such minor differences. Using the emission properties of rice (fluorescence and phosphorescence wavelengths and lifetimes) as the features, several ML models were evaluated. It turned out that the artificial neural network (ANN) worked best after training (n = 284), which exhibited a classification accuracy of 96.4% for the test set (n = 72). For 13 unknown rice samples, the proposed "CTE + ANN" model exhibited an identification accuracy of 84.6%. The universality of the proposed geographical origin tracing approach (CTE + ANN) was further extended to wheat samples whose components are generally similar to rice.