Practical Applications And Related Basic Theory Researches Of Image Moment Methods In The Analysis Of Chemical Spectra

Image moments have many advantages,such as the image description capability,multi-resolution ability and invariance property in translation,scaling and rotation.Owning to these advantages,they can be applied to simultsneous quantitative determination of multiple target compounds in complex samples on be basis of chemical spectra,even in the presence of multiple interferences.Therefore,image moment methods combined with chemical spectra that obtained from different instruments can be regarded as an effective way for quantitative analysis of multiple target compounds in complex samples.The research works in this thesis mainly focused on the practical applications and related basic theories of image moment methods in the analysis of chemical spectra.These research works are organized into the following five chapters?from Chapter 2 to Chapter 6?,following the brief introduction in Chapter 1.Chapter 2 The determination of multiple target components utilizing ¹H NMR spectra combined Tchebichef momentsNMR spectroscopy is a powerful analytical technique.Although a great deal of attention has been paid to the abilities for structural elucidation,its potential for quantitative analysis is also attractive.However,the accurate quantification based on NMR spectra is hampered because of the high degree of overlapping characterizing signals.In the present work,an approach to the simultaneous determination of five target compounds in mixtures by Tchebichef moment method based on the grayscale images of ¹H NMR three-dimensional?3D?spectra with eight gradient levels(range from 3.4 to 26.9 G cm^-1),where existed the different degrees of overlapping ¹H NMR signals was firstly presented.The statistical parameters including correlation coefficient for calibration?R?,leave-one-out cross-validation(R_cv)and prediction?R_p?,root mean squared errors for calibration?RMSE?,leave-one-out cross-validation?RMSECV?and prediction?RMSEP?were calculated to estimate the reliability and accuracy of calibration models,and the predictive ability in practical applications.Moreover,compared with PLS?partial least squares?and N-PLS?N-Way partial least squares?methods,our method is more accurate and reliable,which is expected to be extended as an attractive method to quantify multiple target compounds in complex samples on the basis of grayscale images of ¹H NMR 3D spectra.Chapter 3 The simultaneous quantitative analysis of multiple target metabolites by ¹H NMR self-constructed three-dimensional spectraImage moment is a kind of feature extraction method that can be applied to extract characteristic information on the basis of grayscale images.In general,the conventional 2D?two-dimensional?spectrum for each sample can be obtained when the ¹H NMR is employed.In this chapter,in order to effectively extract the characteristic information of the multiple target compounds from this kind of spectra,and to realize the purpose of accurate quantitative analysis,we propose a self-construction method to obtain self-constructed 3D?SC-3D?spectra from conventional ¹H NMR spectra.Wine consists of several hundred components with different concentrations,and the major components are water,ethanol,glycerol,organic acids,sugars and various ions.NMR is a versatile spectroscopic technology that can be employed as a replacement for time-consuming methods in quality control of wines.However,accurate quantification of target compounds in such complex samples is a difficult task based on conventional ¹H NMR spectra due to some challenges.The Tchebichef moment has been employed to extract important features of the target analytes directly on the basis of the grayscale images of ¹H NMR SC-3D spectra for accurate quantitative analysis of five metabolites in wines.Moreover,the quantitative analysis results of proposed strategy were then compared with that of PLS and iPLS?interval PLS?methods based on the aligned ¹H NMR spectra.The results indicated that the proposed approach can provide accurate and reliable concentration predictions,probably the best results ever achieved using PLS and iPLS methods.Our novel strategy has not only good quantitative analysis performance but also does not require time demanding and laborious multi-step pretreatments.Therefore,it is expected that the proposed method could extend the application of image moment and conventional ¹H NMR spectra in the quantitative analysis of complex samples.Chapter 4 Tchebichef moment method combined with data fusion strategy for the determination of multiple target components in red wine samplesIn the above two chapters,the quantitative analysis purposes have been achieved on the basis of the spectra that obtained from one instrument.However,in some cases,the data obtained from an instrument do not guarantee the satisfactory analytical results.It is likely that,the information content will be enhanced by the combination of data sets coming from different instruments or detectors,and more accurate results will be obtained when the analysis is performed on the basis of the fused information.In the presented work,two data fusion ways?signal fusion and feature fusion?were employed to quantitative determination of nine target compounds in wine samples.In the signal fusion,the SC-3D GC landscapes and SC-3D IR spectra that constructed from original GC and IR signals are combined directly,then the characteristic features were captured from the fused landscapes,and finally the characteristic features were employed for quantitative determination of the nine target compounds.As for the feature fusion,first of all,the features were extracted from SC-3D GC landscapes and SC-3D IR spectra individually,and then concatenated into a single matrix,and finally the fused characteristic features were employed for quantitative determination of the same target compounds.The obtained results indicated that the feature fusion with Tchebichef moments was more effective for determination of the nine target compounds in red wine samples.This study confirms that Tchebichef moment possesses powerful feature extraction capability,moreover,this combination approach is expected to be a promising tool to be used to quantitative determination of other compounds,which are difficult to be determined by individual IR or GC techniques.Chapter 5 The exploration of multi-resolution capability of Tchebichef momentsThe Tchebichef moment method have been successfully applied to quantitative analysis on the basis of HPLC-PAD,LC-MS,¹H NMR,IR and GC 3D?three-dimensional?/SC-3D?self-constructed three-dimensional?spectra,even in the presence of several interferences.It is owing to the multiple advantages of Tchebichef moment,especially the multi-resolution capability that the information of interferences and chemicals can be separated individually,making the quantitative analysis results of the Tchebichef moment method less vulnerable to the interferences.In the present work,a deeper study has been performed to illustrate and investigate the multi-resolution capability.Excitation-Emission matrix?EEM?spectroscopy allows the simultaneous collection of excitation and emission spectra of samples and contains quantitative analytical information which is multidimensional in nature.The spectral information content is increased with EEM,making it easier to accurately quantify multiple components samples.However,serious scatterings and overlapping signals hinder the applications of EEM spectra.In the present work,the multi-resolution capability of Tchebichef moments was investigated in depth on the basis of two EEM data sets?data set 1 consisted of three components and their spectra significantly overlapped;data set 2 included three components with their spectra overlap,and severe scatterings are presented?.The correlation coefficients between the"actual"and"added"Tchebichef momemts obtained from data set 1 demonstrate that the strong overlapped signals can be well resolved within a certain range of moment orders.The results obtained from data set 2 demonstrated that by means of the Tchebichef moments with different orders,the information of chemicals and scatterings can be separated.All these results demonstrated that the different information can be presented by the Tchebichef moments with different orders,which proved effectively the multi-resolution capability of Tchebichef moments.One can end up by saying that Tchebichef moment can resolve the problems of overlapping and scatterings in EEM spectra simultaneously.Chapter 6 Tchebichef-Hermite method:A novel tool for the analysis of three-dimensional spectra and the release of its graphical user interfaceAlthough Tchebichef moment has exhibited excellent quantitative analysis performance,it also has its shortage-the need for the determination of optimum maximal moment orders so as to extract important features?calculate Tchebichef moments?.In the present work,a new method,Tchebichef-Hermite method was proposed to overcome the shortage of Tchebichef moment in the process of selection and optimization of moment orders.The performances of the Tchebichef-Hermite method were tested with two instances for the quantitative analysis of multiple target compounds on the basis of HPLC-PAD and LC-MS 3D spectra,respectively.Experimental results indicate that the Tchebichef-Hermite method not only has the common advantages of these discrete orthogonal moments to deal with some fundamental challenges?such as partially overlapped signals,uncalibrated interferences,peak shifts and baseline drifts?during the analytical process of different kinds of 3D spectra,but also has its unique superiority in information extraction ability that simplify the determination of optimum maximal orders,which makes it much more convenient and efficient in the quantitative analysis process.For the convenience of using the Tchebichef-Hermite method?THM?for quantitative analysis purpose on the basis of 3D landscapes,a user-friendly graphical user interface?THMGUI?for implementing the Tchebichef-Hermite method has been released.It consists of four modules,namely"Import data","Image moment method","Splitting"and"Regression".The THMGUI allows a user to import data and apply the implemented methods in an easy way,in which the characteristic features acquisition,model establishment,and its subsequent application in the analysis of unknown samples are straightforward from the interface.Moreover,the quantitative analysis results including established models,statistical parameters,calibration lines and prediction results are conveniently managed through THMGUI shells and can be presented intuitively.These advantages make the THMGUI useful for less experienced users to adequately process 3D landscapes for quantitative analysis purpose.