Study And Application Of Infrared Spectral Quantitative Analysis Key Algorithms

Quantitative analysis is a key problem in infrared spectra analysis, it's an important method to determine the nature of substance. It has been widely employed in Agriculture,Food, Biology, Pharmacy, Environment and other related fields. Infrared spectral quantitative analysis is a method to estimate concentrations with a prior built quantitative analysis model, which is built with the infrared spectra and corresponding concentrations.Infrared spectra signal correction and partial least square regression are two key parts in quantitative analysis. Efficient and robust infrared spectral signal correction and partial least square regression model will highly increase the accuracy of quantitative analysis.Machine learning and pattern recognition lay a solid foundation for improving the accuracy of quantitative analysis. In this dissertation, machine learning and pattern recognition theory are applied to carry on a deeply research on infrared spectra signal correction and partial least square regression. The main contributions of this dissertation are summarized as follows:1. Infrared spectral sparse orthogonal signal correction model is proposed. Contraposing low accuracy in quantitative analysis of existing orthogonal signal correction model.Sparse orthogonal signal correction model is proposed by introducing the 1norm constraint to direction vector. The orthogonality between removed component and response matrix is enhanced. Results demonstrate that the proposed method can obtain higher quantitative analysis accuracy.2. An Infrared spectral robust model transfer frame based on 1norm and subspace learning is proposed. Existing model transfer methods are sensitive to outliers and noise,which lead to low quantitative analysis accuracy. A robust model transfer frame which utilizes 1norm objective function and subspace learning is proposed. The coefficients of master and slave spectra are employ to calculate transformation matrix. The experimental results indicate that the proposed model is insensitive to outliers and noise and can obtain high quantitative analysis accuracy.3. A new infrared spectral multivariate scatter correction is proposed. Contraposing the poor ability of modeling complex spectra energy fluctuation, which leads to low quantitative analysis accuracy in multivariate scatter correction. A new regularized multivariate scatter correction is proposed. In this model, measured spectra approximates reference spectrum with the weighted sum of nonlinear mapping of obtained spectrum, meanwhile the weight is constraint. Experimental results show a substantial decrease in quantitative analysis errors and it's more powerful to model the complex spectra energy variation than exiting methods.4. Infrared spectral sparse partial least squares regression model based on 2,1norm is proposed. In traditional sparse partial least squares regression model, selected variable are not highly related with response matrix thus it's limited in improving quantitative analysis accuracy. Based on sparse representation and convex optimization theory, A new sparse partial least squares regression model is proposed. In this model, 2,1norm is introduced to constraint direction matrix therefore the selected variables are more highly related with response matrix. The experimental results show that the proposed method obtains higher quantitative analysis accuracy.5. Infrared spectral robust partial least squares model is proposed. Traditional partial least squares model is sensitive to outliers and noise, which lead to low accuracy. Based on sparse representation and convex optimization theory, a new robust partial least squares model which utilize 1norm as objective function to robustify the model is proposed. The experimental results demonstrate that the proposed model is insensitive to outliers and it has higher quantitative analysis accuracy.