Spectral Preprocessing Method And Its Integrated Research

Spectral signals of complex samples are usually disturbed by stray light,noise,baseline drift,and other interference factors,affecting the final qualitative and quantitative analysis.Therefore,it is indeed necessary to eliminate the effects of these factors on the spectral signal before modeling.At present,two kinds of spectral preprocessing selection methods have been developed.One is directly selected according to the characteristics of the spectral signal.The other is to select the preprocessing method according to the modeling functions.Scientific and rational selection methods are still unclear.At the same time,for different spectral data,the optimal preprocessing methods are often distinct.Thus,this thesis investigates different preprocessing methods through a large number of datasets,and a selective ensemble preprocessing method is proposed.The main research achievements contents are shown as follows:1.The necessity of spectral preprocessing and the scientific preprocessing selection methods are discussed.The 10 often used preprocessing methods are in the order of baseline correction,scatter correction,smoothing and scaling to obtain 120 preprocessing methods.120 different preprocessing were performed on the basis of different data and different components of the same data.The results show that the selection of preprocessing method is related to the spectral and related prediction components.Thus,there is no universal preprocessing method.Moreover,the selection of the optimal preprocessing methods should be determined according to the modeling effect of the spectra and the prediction components.2.A selective ensemble preprocessing method is proposed by ensemble technology with preprocessing methods.Firstly,the 10 preprocessing methods are combined in the above-mentioned order to obtain 120 pretreatments and their combination methods,and then the PLS model is established for each preprocessing method.The predicted values of models those are superior to PLS were simply averaged to obtain the final predicted results.Near-infrared spectra of corn,blood,and edible blend oil samples were used to evaluate the performance of the method.Results demonstrate that the selective ensemble preprocessing methods can give comparative or even better results than the best preprocessing method.Accordingly,in the framework of selective ensemble preprocessing,more accurate calibration can be obtained without searching the best preprocessing method.