Data fusion of near-infrared diffuse reflectance spectra and transmittance spectra for the accurate determination of rice flour constituents

The data fusion method effectively fuses multiple complementary inputs for highly accurate analysis. The spectral signals collected by near-infrared diffuse reflectance (NIRr) and diffuse transmission (NIRt) contain various information on the physical structure and chemical composition of the sample. Thus, the data fusion method (for NIRr and NIRt) can be used to further improve the accuracy of the NIR quantitative analysis method. The NIR spectroscopic analysis of protein content (PC), amylose content (AC), and fat content (FC) of rice can be used to select high-quality rice varieties. The data obtained using the NIR spectroscopic analysis method for rice flour were used to optimize NIRr and NIRt data fusion and verify the feasibility of this method to achieve more accurate quantitative analysis. Two types of rice flour spectra, NIRr spectra and NIRt spectra, were processed by different pretreatment methods to obtain high-quality fused spectra. The combinations of different pretreatment methods and spectral ranges were subsequently used for the optimization and calibration of partial least square models. The results reveal that the models of the fused spectra processed by the first derivative [NIRr-NIRt (1 der)] exhibit optimal prediction accuracy. The root mean square errors of prediction (RMSEPs) of the optimal NIRr-NIRt (1 der) PC, AC, and FC models were 0.280, 1.240, and 0.165, respectively, which were lower than those of the NIRr and NIRt models. The results show that the fusion of NIRr and NIRt data can achieve accurate detection of rice flour constituents, indicating the method has potential for further development and application.