Multi-frequency microwave dielectric properties-based method coupled with SPA-PLSDA algorithm for rapid discrimination of grain mildew

Abstract

To achieve in-situ monitoring of the occurrence of grain mildew and ensure food safety, this study took paddy grains as the object and carried out the discrimination of paddy mildew based on microwave dielectric properties. The multi-frequency swept measurement technique was used to acquire the dielectric constant (DC) and dielectric loss factor (DLF) spectra (2.00–10.00 GHz) of healthy samples and samples with different moldy paddy content. To choose the most effective frequencies, 20 frequency subsets (DC subset 1–10, DLF subset 1–10) were generated by an algorithm coupled with the successive projections algorithm and partial least-squares discriminant analysis. Afterward, four key frequencies were determined from the 100 pairwise combinations of the two types of frequency subsets by exhaustive method. Dielectric properties at key frequencies and sample thickness were used as the input variables to establish the discriminating model for paddy mildew. The established microwave dielectric properties-based model achieved 100% accuracy in distinguishing healthy and moldy samples, showing perfect discriminant validity. Moreover, only 4.4% of those samples whose MPC was at a low level (≤30%) were misclassified by the model, and the discrimination model achieved 97.29% overall accuracy. The results of this study should encourage future research on dielectric-based mildew detection in food processing and agriculture-related industries.