Spline transform partial least squares penalty function integration method for reconstruction and quantitative analysis of spectral data_0.docVIP

 PAGE \* MERGEFORMAT 18 Spline transform partial least squares penalty function integration method for reconstruction and quantitative analysis of spectral data [Abstract] for high-dimensional spectral data show a small sample of the functional data (Functional data characteristics, and nature of parameters and the nonlinear relationship exists between variables of serious collinearity, using the penalty function spline partial least transform integration least squares regression of new technologies. It first the cubic B-spline-based nonlinear spectral transform linear data reconstruction and will be followed by reconstruction of the new spectral penalty function matrix partial least squares (Penalized PLS build its nature of the correction between parametric model, in which the smoothness penalty function in the optimization of factors determined by the cross-validation in order to control the model fitting precision. Finally, the sample moisture content of wheat near-infrared spectroscopy revealed that the technical reconstruction of spectral data sound, noise is obvious, and effective solution to high-dimensional small sample of the over-fitting and collinearity among variables, and the root mean square error of prediction set (RMSEP was 0.1808 percent, the method of nonlinear model predictive ability has been corrected improved significantly. [Keywords:] spline function, partial least squares, rough punishment, near infrared spectroscopy, quantitative analysis, wheat 1 Introduction Modern spectral analysis of its speed, reproducibility, low cost, do not consume the sample, and so easy to implement online analysis is widely applied. The spectra Chemometrics infrared spectroscopy is an important modern part of it through a multi- multivariate calibration techniques to the data (sample spectrum and the nature of processing parameters to obtain accurate results [1,2]. taking into account the near-infrared spectral data were generally more variable