Study On The Selection Of Near-Infrared Spectral Variables Based On Model Population Analysis

Near infrared spectroscopy is a fast and efficient detection technology,which has been widely used in various fields in recent years,and plays an increasingly important role in the field of quality analysis of agricultural products.However,there is a very important problem in the analysis of near infrared spectroscopy,which is that there is a large amount of redundant information in the spectrum,and the characteristic absorbing region of the component is not clear.Variable selection can improve the quality of the model,and this paper mainly studies the variable selection method based on model cluster analysis(MPA).Because the existing variable selection method is mostly based on a modeling idea,the number of samples collected is not enough to express the overall information,which leads to the model being fitted or nonstandard.Therefore,MPA maximizes the use of existing samples and achieves maximum information acquisition through random sampling.Bootstrapping soft shrinkage(BOSS)and Monte Carlo-variable combination population analysis(MC-VCPA)are all combinations of MPA near infrared spectral variable selection method.BOSS combines soft shrinkage strategy and the weighted guided sampling(WBS)method and provides a method to extract information from the model regression coefficient(RC).MC-VCPA randomly sampled the sample space through the Monte Carlo sampling(MCS)method,and combined with VCPA to select the characteristic variables of the variable space of different sample subsets.In this paper,we use the two methods of variable selection based on MPA to establish the predictive model by partial least squares(PLS)and predict the forecast set samples.Contrast UVE and GA variable selection algorithm,such as BOSS for beer yeast concentration prediction,compared with the full spectrum of PLS model predicted values of root mean square error(RMSEP)dropped from 0.5348 to 0.1565,and the forecast precision is improved by 70%.MC-VCPA is prediction of wheat protein content was reduced from 0.5096 to 0.3295,and the prediction accuracy was increased by 35% compared with that of total spectrum PLS.The results show that both BOSS and MC-VCPA have significantly improved the prediction accuracy of the model by quantitative analysis of near infrared spectral data analysis.