Application And Study Of Consensus Modeling Method In Spectral Analysis

Spectral analysis technology has been successfully applied in many fields such as agriculture and food due to its simple operation,multi-element or compounds determination simultaneous and high sensitivity. Single-model modeling is a commonly used technique in spectral analysis, namely to construct an optimal model by the training samples to predict unknown samples. However, when there are some complicated situations such as a limited number of training samples, it is difficult to meet the requirements for the results of the single-model modeling method. Consensus modeling method has no tendency to the quality and size of the data set and stronger robustness for the uneven data set, thus it is able to compensate for the inadequacy of the single-model modeling method. Consensus modeling refers to form a consensus model by combining a series of member models, so as to achieve the improvement of the prediction accuracy and generalization performance of the model. This paper studies the application of consensus modeling method in quantitative and qualitative analysis of near-infrared spectra and laser-induced breakdown spectra. The main contents are as followings:1. Introduced the development history of spectroscopic research, the basic principle of spectral analysis and the application of chemometrics method, the commonly used single-model modeling methods and consensus modeling methods in spectral analysis and their applications.2. Studied and analyzed the deficiency of single-model method of SPA-MLR. In order to fully exploit and extract the useful information contained in spectra, a new multi-regression model consensus modeling method named consensus SPA-MLR(C-SPA-MLR) was proposed. In this method, SPA-MLR is used to construct several member models with different subsets of variables, which are selected from the remaining variables iteratively,then the member models that have better prediction performance are combined to form a consensus model for further prediction of unknown samples. The C-SPA-MLR method was evaluated by analyzing the near infrared(NIR) spectra of corn. The results of this method showed a better prediction performance compared with the SPA-MLR and full-spectra PLS methods.3. Studied the feature information extraction method in the pattern recognition process, a classification method that combines wavelet packet transform and multi-classifier consensus modeling method was proposed.First, the feature information of signal is extracted by wavelet packet transform, in which wavelet coefficients obtained from the decomposition that can reflect different feature information are taken as characteristic variables directly, then these variables are used to establish qualitative calibration member models(namely member classifiers), finally the multi-classifier consensus modeling method is used to combine them together to form a consensus classifier for classification research. The proposed method was evaluated by analyzing the laser-induced breakdown spectra of heavy metal-contaminated Tegillarca granosa. The results showed that the proposed method improved the classification accuracy of the model, and had better performance than other two recognition methods which based on the full spectra and the optimal wavelet packet coefficients.