论文部分内容阅读
For the regression model for quantitative analysis based on near infrared spectroscopy,the optimization method of measurement model can improve the predictive accuracy while saving the involved data amount and the time of calculation.In this work,five methods,including orthogonal signal correction (OSC),interval partial Least-Squares (iPLS),variable importance on the projection (VIP),genetic algorithm (GA) and uninformative variable elimination (UVE),were applied into optimizing the model for measurement of sugar content in chestnut which was established on the near infrared spectra of 185 chestnut samples in the wavelength range from 833 nm to 2500 nm.The OSC algorithm can increase the correlation coefficient (R) of model for validation set to 0.8961 and decrease the RMSEP to 0.63.The iPLS method and VIP algorithm can function as well as the original model only with half of the variables,which means the involved data amount and calculating time would be reduced sharply.The GA method and UVE algorithm were able to establish a model only using less than 17% of the variables,however,the prediction accuracy were not increased.These results showed that the algorithms of model optimization had the potential to improve the performance of the measurement model for sugar content in chestnut based on near infrared spectroscopy.