| Abstract: | Fat content in rice is one of the most important nutritional quality properties. But the chemical analysis of fat content is time‐consuming and costly and could result in poor reproduction between replicates. Near‐infrared spectroscopy (NIRS) can solve those problems by providing a rapid, nondestructive, and quantitative analysis. Based on the NIRS technique and partial least squares (PLS) algorithm, four calibration models were established to quantitatively analyze fat content in brown rice grain and flour and milled rice grain and flour with 248 representative samples. The determination coefficients (R2) of these calibration models were 0.79, 0.84, 0.89, and 0.91, respectively, with the corresponding root mean square errors 0.16, 0.14, 0.09, and 0.08%. The R2 were 0.73, 0.81, 0.81, and 0.89 with the corresponding root mean square errors 0.17, 0.15, 0.12, and 0.09%, respectively, in cross validation. The R2 were 0.62, 0.80, 0.81, and 0.87, respectively, with the root mean square errors 0.25, 0.31, 0.28, and 0.30% in external validation. These results indicate that the method of NIRS has relatively high accuracy in the prediction of rice fat content. The four calibration models established in the present study should be useful for nutrient quality improvement in rice breeding. |
