Effect Of Moisture And Particle Size On Prediction Performance Of Maize And Wheat Fourier Near Infrared Spectroscopy Model

To study the fitting particle size by near infrared spectroscopy and the predict sample fitting moisture content for the model which have the background of some moisture contain and particle size. Use Fourier near infrared spectroscopy and Partial Least Squares, use 73 different hybrids maize samples and 89 different white wheat samples as materials. All samples were crashed 20, 40, 60 mesh respectively, and to establish the crude protein(CP), ether extract(EE), ash, NDF , ADF models in each particle size, to study effect of particle size on predict-performance ; Use the same model to predict some samples which dried different times by oven, to study the effect of moisture on predict-performance, the result indicated:(1) The study showed the all models of maize and wheat have the best predict- performance in 40 mesh size except the ash in wheat model, which better as 60 mesh size. In the 40 mesh maize models, the CP , EE , ash , NDF , ADF calibration coefficients of determination (R_cal²) value are 94.02, 95.60, 91.88, 95.40, 86.19 respectively, the external validation coefficients of determination (R_val²) value are 91.85, 88.07, 76.12, 90.56, 85.60 respectively, the RSD value are 2.70%, 4.5%, 3.2%, 3.09%, 1.75% respectively. Coefficients of correlation between NIRS predicted values and chemical values were 0.96(cp) - 0.87(ash). In the 40 mesh wheat models the calibration coefficients of determination (R_cal²) value are 96.86, 89, 89.21, 89.30, 88.99 respectively, the external validation coefficients of determination(R_val²) value are 98.82, 43.9, 72.48, 88.34, 84.79 respectively, the RSD value are 1.22%, 6.32%, 5.20%, 3.05%, 4.15% respectively. Coefficients of correlation between NIRS predicted values and chemical values were 0.99(cp)-0.72(EE).(2) when the models have the background of moisture ranged from 9% -14% and 40 mesh size, the fitting moisture ranged from 11.21% to 12.85% for maize samples, and the fitting moisture ranged from 11.95% to 13.60% for wheat samples which used to predict cp, NDF and ADF.(3)There has an additional effect on predict-performance between moisture and particle size. In the 20 mesh size models, reduced the moisture contain can improve the model's predict-performance. For the cp model, the fitting moisture ranged from 10.50%-11.95% for maize samples, and 11.30% -12.44% for wheat samples. For the ADF model, the fitting moisture ranged from 10.75% to 11.66% for wheat samples. In the 60 mesh size models, reduced the moisture contain also worsened the model's predict -performance.