Research On Spectral Correction Methods And Its Application In Soil Detection

Near-infrared spectroscopy analysis is a fast,simple and green analytical technique.Diffuse,transmission reflectance mode and so on is used in the directly rapid measurement of solid sample.Meanwhile it also poses a challenge in methodology.Because it is necessary to overcome the interference of spectral baseline drift and tilt caused by light scattering.Multiplicative scatter correction?MSC?,standard normal variate transformation?SNV?,and Savitzky-Golay smoothing?SG?are important spectral correction methods that can be used to correct spectral interference due to scattering,temperature fluctuations,etc.It can also be used for model transfer between different instruments.MSC assumes that the spectrum of each sample is linearly related to the average spectrum in the calibration set.But for the wider spectrum,the assumption of overall linearization is difficult to establish.Piece-wise multiplicative scatter correction?PMSC?method proposed a local correction window with movable p+1+q patterns,achieving the piecewise linear correction of the spectrum.Through several examples,the prediction effect of the PMSC method is obviously better than that of the MSC method.However,subjected to the computer operation speed at that time,the PMSC did not achieve global optimization.On the other hand,the spectral correction parameters for different wavebands are usually different.Due to the complexity of the algorithm,the simultaneously interactive optimization method for waveband selection and spectral correction has not yet been reported.The extension methods in two respects were carried out in this study:?1?An optimization platform of PMSC method based on partial least squares?PLS?models,to achieve global optimization of the correction window pattern;?2?Based on four spectral correction methods of SG,SNV,MSC and PMSC,combined with moving window PLS?MW-PLS?,four integrated optimization methods including MW-SG-PLS,MW-SNV-PLS,MW-MSC-PLS and MW-PMSC-PLS were proposed,realizing the simultaneously interactive optimization between waveband selection and spectral correction.Taking the near-infrared analysis of soil organic matter as an example,several proposed methods were verified.The main results are as follows:?1?PMSC-PLS method:?1?PMSC-PLS model for global optimization algorithm platform was established,to implement global screening of correction windows.The root mean square error and correlation coefficient in validation set of the optimal model(RMSEP_V,R_P,V)were 0.184g·kg^-1 and 0.975,respectively;which were significantly better than the MSC-PLS model.?2?Research found that the correction window of PMSC-PLS model is divided into three categories:Excessive,moderate and insufficient correction.In this case,the window width is too small?m<61?for excessive correction;the window width is too large?m>800?for insufficient correction;and the range of appropriate correction window width is in the range of 61 and 800.?2?Interactive optimization method:Using the overall primary screening parameter optimization strategy,the interactive optimization operations were implemented for the four integration methods:?1?The RMSEP_V and R_P,V of the optimal MW-SG-PLS model are 0.187 g·kg^-1 and 0.969,respectively,which are superior to SG-PLS and MW-PLS models;?2?The RMSEP_V and R_P,V of the optimal MW-SNV-PLS model are 0.186 g·kg^-1 and 0.970,respectively,which are superior to SNV-PLS and MW-PLS models;?3?The RMSEP_V and R_P,V of the optimal MW-MSC-PLS model are0.188 g·kg^-1 and 0.970,respectively,which are superior to MSC-PLS and MW-PLS models;?4?The RMSEP_V and R_P,V of the optimal MW-PMSC-PLS model with symmetric correction window are 0.177 g·kg^-1 and 0.971,respectively,which are superior to PMSC-PLS with symmetric correction window and MW-PLS models.Therefore,the interactive optimization of spectral correction and waveband selection is superior to both the separated spectral correction and the separated waveband selection.?3?Among the above methods,the MW-PMSC-PLS model with correction window optimization and waveband selection simultaneously has the best prediction performance.In this research,the global optimization of the MSC with alterable correction window was implemented.The interactive optimization of spectral correction and waveband selection was achieved as well.These methods extend the application of near-infrared spectroscopy,which provides important references for the near-infrared spectroscopy analysis of complex systems.