Chemical High-dimensional Pattern Recognition Methods And Their Applications In The Quality Control Of Food And Traditional Chinese Medicine

With the rapid development o f analysis instruments,the data obtained by analytical chemists has also changed from traditional scalar data or vector data to more complex high-dimensional data.These data co ntain more physicochemical i nformation,but are also more diff icult to process.How to obtain useful information from these complex measurement data and apply it in practice has become a new pursuit of analytical chemists,and chemometrics has emerged as the times require.As the cor e content of chemometrics,chemica l multivariate calibration and resolution and chemical pattern recognition are the key to solving this problem.By focusing on this problem,the author of this paper carefully analyzes the urgen t problems and research hots pots in the fields of food and med icine at home an d abroad,according to the characteristics of the data generated by different measurin g instruments,organically combines the chemometrics algorithm,and applies it It is used to solve problems such as origin traceability and adulteration d iscrimination in the fields of food and medicine.Part Ⅰ: Excitation-emission matrix fluorescence combined with chemical high-dimensional pattern recognition for food quality c ontrol(Chapter 2-Chapter 4)In Chapter 2,we combined excit ation-emission matrix three-dimensional fluorescence fingerprints with chemometrics high-dimensional pattern recognition methods Partial least squares-discriminant analysis(PLS-DA)and Principal component analysis-linear discriminant analysis(PCA-LDA)to discriminate rock tea of different tree s pecies and planting areas.T he classification models can obtai n high classification accuracy,and the classification accuracy of the training set is hig her than 99.12%.In addition,the adulteration discriminant model built by PLS-DA can also identify pure beef samples from four type s of adulterated samples of niulan keng rougui(niu rou)(shuixian + niurou,rougui + niurou,pinzhong + niurou,pure niurou),the classification accuracy of the training set is 100%,the classification accuracys of the test set and prediction set is higher than 90%,no pure niurou samples a re misclassified,and no adulterated niurou samples are misclassified as pure niurou.After that,the accurate quantitative analysis of the adulteration level was also completed based on the PLS regression model.The above results all show that the propose d methods can cl assify different types of rock tea,and can also complete the accurate ide ntification of adulterated niurou and rough prediction of the adulteration level.It is also expected to be popularized in the identification of adulteration in the f ield of food and medicine.Compared to excitation-emission matrix three-dimensional fluore scence,four-dimensional fluorescence data may provide more abundant sample information,which can impro ve classification accuracy.In Chapter 3,we proposed four-dimensional fluores cence data combined with a multi-way chemometric algorithms to characteriz e and classify Chinese lager beers from different manufacturers.Undiluted and diluted beer samples exhi bited different fluorescent fingerprints due to the matrix eff ects.In this wo rk,dilution level is added as an additional dimensionality for the first time to construct four-way sample-excitation-emission-dilution level data array.The classification mode ls are built by the proposed four-dimensional fluorescence dat a to improve classification accuracy.Firstly,excitation-emission matrix three-dimensional fluorescence of beer with different dilution were collected and combined into the four-dimensional flu orescence data.The construc ted three-way and four-way data arrays were decomp osed using three-way and four-way parallel factor analysis(PARAFAC),resp ectively,to further realize beer characterization and feature extraction.Based on the features extract ed in different ways,four b eer classification strategies are proposed and stu died.Strategy 1 builds a discriminant model based on the three-way PARAFAC scores of undiluted beer data;strategy 2 builds a discriminant model based on the three-way PARAFAC scores of diluted beer data;strategy 3 builds a discriminant m odel based on th e fusion data of the above three-way PARAFAC scores;strategy 4 builds a discriminant model based on four-way PARAFAC scores In each strategy,three supervised classification met hods,linear discriminant an alysis(LDA),partial least squares discriminant a nalysis(PLS-DA),and k-Nearest Neighbors(k-NN),were used to construct d iscriminant models.The comparative analysis shows that the classification results of the PARAFAC-data fusion-k-NN method in strategy 3 and the four-way PARAFAC-k-NN method in strate gy 4 are the best,and the correct classification rate of cross-validation are all 95.8%,the correct classification rate for the test set are all 91.7%.This is also the first t ime to construct a sample-excitation-emission-dilution four-dimensional data a rray to solve the classification problem.As craft beer replaces industria l beer as the new favorites of China’s beer market,counterfeiting of craft beer has become a common pro blem.In Chapter 4,the ingenious strategies proposed in chapt er 3 were applie d to achieve the authentication of industry and craft beers in the Chinese market.Excitation-emission matrix fluorescence spectroscopy technique was used to obtain high-dimensional data of beer samples wit h 3 dilution levels(the content o f beer in water=3%,50% and 100%).Alternating trilinear decomposition(ATLD)and alternating quadrilinear decomposition(AQLD)algorithms were used to characterize beer samples.Three chemical pattern recognition methods including PLS-DA,PCA-LDA and random forest(RF)were used to handle two classification tasks with different aims,inclu ding the classification of i ndustry beer and craft beer(Case 1)and the class ification of industry beer and craft beers of different brands(Case 2).For case 1,the correct classification rates of the cross-validation,training set,test set and predicti on set are all above 100% obtained by the classification metho ds based on four-way data arrays.For case 2,the results obtained by PLS-DA based on four-way data are excellent,the correct classification rates of cross-validation,training set,test set an d prediction set are 90.3%,100.0%,100% and 88.9%,respective ly.This study showed that four-way data coupled with PLS-DA could be a good choice for all beer classification tasks.The proposed method is expected to safeguard the interests of consumers and manufacturers,and it can b e extended to the authentication o f other beverage s.Part Ⅱ: Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry combined with chemical pattern recognition for origin traceability of traditional Chinese medicine(Chapter 5)In Chapter 5,geographi cal origin and authenticity are two core factors to promote the development of Traditional Chinese medicine(TCM)herbs perception in terms of quality and price.Therefore,they are important to both sellers and consumers.Herein,we propose an efficient,accurate method for discrimination of genuine and non-authentic producing areas of TCM by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry(MALDI-TOF MS).Take Atract ylodes macrocephala Koidz(A MK)of compositae as an example,the MALDI-TOF MS spectra data of 120 AMK samples aided by principal component analysis-linear discriminant analysis(PCA-LDA),partial least squares discriminant analysis(PLS-DA)and random fore st(RF)successfully differe ntiated Zhejiang province,Anhui p rovince and Huna n province AMK according to their geographical location of origin.The correct classification rates of test set were above 93.3%.Furthermore,5 recollected AMK samples were used to verify the performance o f the classification models.The outcome of this s tudy can be a good resource in building a database for AMK.The combined utility of MALDI-TOF MS and chemometrics is expected to be expanded and applied to the origin traceabilit y of other TCMs.Part Ⅲ: Several data fusio n strategies com bined with chemical pattern recognition for origin traceability of traditional Chinese medicine(Chapter 6)In Chapter 6,Geographical origin has great influence on the quality o f traditional Chinese medici ne.This work reported an applicat ion of geographi cal origin traceability of Atractylodes macrocephala Koidz.based on chemometrics classification methods combined with data fusion of synchronous fluorescence spectroscopy and su rface-enhanced Raman spectro scopy.Firstly,the samples were tested by synchro nous fluorescence and surface enhanced Raman scattering spectroscopy,and then the two types of data were fused by the low-level data fusion strategy and the middle-level data fu sion strategy,respectively.The low-level data fusion strateg y combines the t wo types of spectral data directly,while the middle-level data fusion first performs principal component analysis on the two types of spectral data,and then combines the obtain ed principal components.Fin ally,the classification model was built by princi pal component analysis-linear discriminant analysis,partial least squares discriminant analysis and random forest.The cross-validation showed that the correct classification ra te could achieve 91.0% for p artial least squares discriminant analysis based o n low-level data fusion and the classification model based on low-level data fusion could achieve accurate classification of 14 new Atractylodes macrocephala Koidz.samples.The results demonstrated that the synchronous fluorescence spectro scopy and surface-enhanced Raman spectroscopy complemented each other,and better classification results can be obtained based on data fusion strategy.