Effect of spectrum preprocessing methods on the rapid detection of ash content in flour by the portable NIR spectrometer

In order to carry out the spot rapid analysis of ash content in flour which based on the portable NIR spectrometer, 125 samples of flour was scanned and some quantitative analytical modelswas built which based on the partial least squares ( PLS) analysis after various spectrum preprocessing methods. The experiment discussed the effect of six spectrum preprocessing methods including baseline, normalize, Savitsky-Golay convolution smoothing method, derivation, standard normal variate correction and multiplicative scatter correction and their combination on modeling, and analyzed various influencing factors in the process of experiment.The results showed that the best spectrum preprocessing method was the combination of normalize and Savitsky-Golay convolution smoothing method. The correlation coefficient of calibration ( R_c²) of the best corresponding model was 0.947.The Cross validation coefficient R² vwas 0.896. The root mean square error RMSEC was 0.026. The root mean square error RMSECV was 0.037.The root mean square error of prediction set model RMSEP was 0.026. The correlation coefficient of calibration ( R_c²) of the model without spectrum preprocessing method was 0.873.The Cross validation coefficient R² vwas 0.832.The root mean square error RMSEC was 0.044.The root mean square error RMSECV was 0.051.The root mean square error of prediction set model RMSEP was 0.056.Compared with the model without spectrum preprocessing methods, the best model strongly improves the prediction accuracy and the robustness of the model.