Study On The Model Establishment And Transfer Of Near Infrared Spectroscopy

Abstract:The research of model establishment and transfer is important to near infrared spectroscopy.Therefore,the RobustICA-PSO-ELM method which is used to model is proposed and applied to quantitative and qualitative analysis.Model transfer of sugar model in different tree shapes and the rice protein powder in different instruments are studied.The following aspects are involved in the paper:(1)The robust independent component analysis is applied to the spectral feature extraction for the large amount of spectral data.Because of the fast convergence rate,this method can select the best step length and any independent component that is not zero to ensure to extract independent component which contains sufficient useful information,improve the robust.(2)In order to solve the weak problem of model and the low accuracy,a RobustICA-PSO-ELM model is established.The model can extract useful information in the spectrum,improve the prediction ability and accuracy.In the qualitative analysis,based on the RobustICA-PSO-ELM method,the model of cuiguan pear maturity identification was established and the identification of cuiguan pear is raised to 95.833%.In the quantitative analysis,a model of rice protein powder was established based on the RobustICA-PSO-ELM.The RMSEP and R~2 of water,protein,fat were greatly improved.(3)The model transfer of cuiguan pear sugar model in different tree shapes and the rice protein powder in different instruments was studied.The P type cuiguan pear PLS model was established.Two S/B algorithms were used and RMSEP dropped from 8.21 to 0.704 and 0.535 respectively.PLS model of rice protein powder was set up.Comparing the different model transfer methods,the DS algorithm was found to be the best.