Rapid Quality Discrimination of Rice during the Mildew Process Based on Multi-Source Information Fusion

In this study, a rapid and non-destructive method was proposed for the quality discrimination of rice at different stages during the early mildew process. The change in volatile organic compounds (VOCs) during the mildew process was analyzed using headspace solid-phase microextraction-gas chromatography-mass spectrometry (HS-SPME-GC-MS). Meanwhile, Fourier transform infrared spectroscopy (FTIR) was employed to monitor the structural change of starch. Following feature variable selection, a feature-level fusion method was applied for data fusion of GC-MS and FTIR. Partial least square-discriminant analysis (OPLS-DA) was used to establish a discriminant model for determining the quality of rice during the early mildew process. Cluster analysis performed on physicochemical indicators categorized the mildew process into three stages. Based on total mold counts, rice became moldy on the 22th day. The model developed through feature variable fusion of GC-MS and FTIR data proved to be the most effective in differentiating the quality of rice at different stages of mildew, which was successfully clustered into 7 stages. The model demonstrated a goodness of fit (R² = 0.95) and a goodness of prediction (Q² = 0.86). In conclusion, data fusion of GC-MS and FTIR could accurately discriminate the quality of rice during the early mildew process, thereby providing the basis for the rapid and non-destructive inspection of rice quality during the early mildew process.