Establishment and Verification of Near-infrared Spectral Prediction Model for Fat Content of Xanthoceras Sorbifolia

In order to realize the nondestructive and rapid detection of fat content of Xanthoceras sorbifolia and meet the screening of breeding materials and industrial processing requirements of X.sorbifolia, 46 X.sorbifolia were selected as the standard sample set, the results showed that the fat content of 46 apricot kernel kernels was 49.38%~68.98% an average content of 61.62%. The fat content of the seed kernel was determined by the Soxhlet extraction method, and the spectral data of the sample was collected by the near-infrared spectroscopy (NIRS) technology, and the Unscrambler software was used to construct the NIRS prediction model of X.sorbifolia fat content by the partial least squares (PLS) method. The results showed that the regression curve R-Square (determination coefficient) of the model was 0.9856, and the RMSE (standard error) was 0.4149, which could be used for effective prediction. At the same time, 32 X.sorbifolia samples not participating in the modeling were selected as validation materials to further carry out external test on the prediction effect of the model. The results showed that the external test regression curve R-Square was 0.9014, RMSE was 0.8259, and the predicted value of fat content was in good agreement with the chemical value.