Rapid Identification Of Silicate Mineral Medicine Based On XRD、NIR Spectroscopy And SVM Algorithm

Thispaperdiscussestherapidanalysisofeightsilicate minerals（Taculm,Actinolitum,Halloysitum Album,Halloysitum Rubrum,Chloriti Lapis,Vermuculitum,Lapis Micae Aureus and Muscovite）based onX-ray diffraction technique（XRD）,near infrared spectroscopy（NIRS）combined with support vector machine（SVM）artificial intelligence algorithm.Since these silicate mineral medicine are poorly soluble in water,general acid or alkali solutions,and remian undecomposed until the temperature exceeds 700℃,it is more difficult and complex to analyze,these mineral medicine using conventional physicochemical method,In addtion,as some of these mineral mechine contain manytypes of positive ions,conventional physicochemical analysis,which is typically represented by cationic color reaction,will show lower specificity to them.Therefore,advanced technologies such as XRD and near infrared spectroscopy（NIR）were adopted for rapid analysis of the 8 silicate mineral medicine in this paper.Beside,conventional chemometrics algorithms such as distance discriminant analysis,principal component analysis（PCA）,PLS were also employed to realize NIR-based rapid identification and quantitative analysis of these mineral medicine.On this basis,this study explores the application of newly-emerging artificial intelligence algorithms such as SVM classification algorithm and regression algorithm,which increases the accuracy of NIR-based rapid identification and provides reference for mineral medicine analysis.First,the traditional morphological identification,physicochemical identificati on and XRD technique were combined to compare character,physicochemical and phase information of each sample and to identify the origin sources of the 146batches of silicate mineral medicines collected on the market.Preliminary judgments and objective analysis were conducted for the138 batches of authentic silicate mineral medicines.Through the powder X-ray diffraction technical analysis,it was confirmed that the main composition of Talc was talcum mine,with a small amount of dolomite,serpentine-chlorite,quartz,etc.The X-ray diffraction pattern showed strong characteristic diffraction peak of talc at d=0.932nm;the main composition of Actinolitum was tremolite,which was accompanied by paragenetic mineral actinolite,and a small amount of chlorite,talc,quartz,etc.The XRD pattern showed strong characteristic diffraction peak of tremolite at d=0.905nm;the main component of Halloysitum Album was kaolin（kaolinite and halloysite）,which was accompanied by illite,and a small amount of talc,quartz etc.The XRD pattern showed strong characteristic diffraction peak of halloysite at d=1.00nm,and strong characteristic diffraction peak of kaolinite at d=0.714nm,with peak shape dispersion at 0.71¹.01nm.The main composition of Halloysitum Rubrum was halloysite,and its XRD pattern showed strong characteristic diffraction peak of halloysite at d=1.00nm.The main composition of Vermiculitum was phlogopite which was accompanied by a small amount of vermiculite,feldspar.The XRD pattern showed characteristic diffraction peak of phlogopite at d=0.994nm,with different drgrees of peak shape dispersion in different degrees at the position of0.75¹.05nm.ThemaincompositionofLapismicaeaureumwas vermiculite-bearing biotite,which was accompanied by vermiculite,biotite and a small amount of amphibole.The XRD pattern showed significant obvious characteristic diffraction peaks of vermiculite-bearing biotite at d=0.496,1.19 nm,with different drgrees of peak shape dispersion in different degrees at the position d=1.0¹.1nm.The main composition of Chloriti Lapis was biotite,which was accompanied by a small amount of feldspar,chlorite and quartz.The XRD pattern showed strong characteristic diffraction peaks of biotite at d=0.263,0.335,and 1.00nm.The main composition of Muscovite was muscovite ore,which was accompanied by a small amount of chlorite and albite.The XRD pattern showed strong characteristic diffraction peaks of muscovite at d=0.332 and 0.994 nm.Based on the results of these studies,it could be known that the above eight mineral medicines or mineral powder.can be accurately identified and distinguished by XRD technique..This study provides the basis for subsequent research,and enriches the research content of mineral medicine identification by XRD technique.Based on accurate identification of the 8 kind of medicinal materials,the NIRS qualitative rapid identification models for the 138 batches of silicate minerals with identified sources were established based on distance discrimination method and SVM intelligent learning machine classification method,,and then established qualitative methods were evaluated and compared.The INDENT module of OPUS software and SVM classifier of Matlab software were used to extract the feature information of spectroscopic data of the 138 batches of authentic samples.The model parameters were optimized through spectral pre-processing,characteristic spectrum segment selection,algorithm optimization,and good classification effect was obtained.By using OPUS software INDENT module,selecting Euclidean distance as the evaluation index of spectrum similarity,establishing the identification mode of secondary sub-libraries,choosing the first derivative+vector normalization method as pre-processing method for primary selection,and spectral region as12000⁴500cm^-1,the three samples such as talc,muscovite and actinolitum could be accurately identified.For other 5 kinds of samples that were not fully identified.Two secondary sub-librariesAandBwerecreatedforfutheridentification,For A-library,pre-processing method was vector normalization method.The spectral region was 10742⁹724,7305⁶842 and 5625⁴000cm^-1,The standard algorithm,was adopted,which realized classification accuracy of halloysitum album,Halloysitum Rubrum and chloriti lapis as high as 93%.For B-library,the pre-processing method used for B-library was first derivative+vector normalization method,the spectral region was 7421⁶879 cm^-1and factorization method was adopted,which realized classification accuracy of vermiculitum,Lapis Micae Aureus as high as 90%.The combined application of main library and its sub-libraries for qualitative identification of these eight medicinal materials realized good classification result with identification accuracy reaching 97%.In addition,a qualitative model for the 138 batchs of samples was established by SVM classification method of Matlab software.The first derivative method was adopted as the pre-processing method,the spectral region was 12000⁴500cm^-1.The six principal components were extracted after principal component analysis and dimension reduction.The classification accuracy of the model established by SVM algorithm reached 100%.The model’s predictive ability was evaluated by test set’s sample,with classification accuracy rate reaching 100%,indicating the feasibility of SVM algorithm in qualitative identification of the 8 silicate mineral medicines,as well as the superiority of SVM intelligent learning machine classifier in small sample modeling than distance discrimination-based traditional pattern recognition method.In this paper,the quality evaluation of the most commonly used mineral talcum in silicate mineral drugs was conducted.We explored the NIRS-based rapid quantitative analysis of MgO content in talc mineral medicine.Fisrt,a large number of representative talc samples were added.Then,various data processing methods were compared according to the MgO content in talc measured by EDTA titration as a reference value and eventually,a better near-infrared analysis model was determined,The verification effect of the models was comprehensively compared.The first four principal component scores（obtained by reducing the dimensions of characteristic segment with Partial Least Squares（PLS））were input variables of support vector machine,The PLS-SVM model constructed by grid optimization algorithm was selected was the optimal talc near-infrared quantitative model.By using PLS-SVM model,MgO content in talc was determined in the range of17.422-33.22%,Root Mean Square Error of Cross Validation（RMSECV）was2.2127,Root Mean Square Error of Calibration（RMSEC）was 0.6057,and Root Mean Square Error of Prediction（RMSEP）was 1.2901.The PLS-SVM model has high model accuracy and strong predictive ability,wich can be used for rapid prediction of magnesium oxide content in talc.1)Basic data for XRD identification of 8 silicate mineral medicine were established.First,XRD method was adopted to identify the sources of 8 silicate mineral medicine and to determine the main components and characteristic XRD spectrum data of each mineral medicine.This data provides basis for the XRD identification of 8 silicate mineral medicine and medicine powder,and enriches the research content of XRD identification of mineral medicine.2)The NIRS-based quantitative analysis model of 8 silicate mineral medicine was established.On the basis of analyzing the sources of 8 mineral medicine with XRD method,,the secondary sub-library identification model was established using INDENT module of OPUS software with Euclidean distance model as spectral similarity evaluation index,Based on this eastalished model,talc,mica and actinolite samples were identified accuratelyFor the other 5 types of samples that were not successfully identified,another two secondary sub-libraries（sub-library A and B）were established.By using processing methods that are respectively corresponding to sub-library A and B and combining main library and sub-libraries,the identification accuracy rate for the 8 mineral medicine reached97%.The qualitative model established based on SVM classification method of Matlab software can achieve 100%accuracy in both classification and identification for 138 batches of medicine samples.This indicates SVM-based intelligent learning machine classifier is more effective than the pattern recognition method based on distance discrimination.3)The quantitative analysis model of MgO content in silicate mineral talcum was established with talc as an example.This model preliminarily demonstrated the feasibility of NIRStechnology in rapid quantitative analysis of silicate mineral medicines.By using the intelligent PLS-SVM algorithm,the best talc near infrared quantitative model was established,with the RMSECV of2.2127,RMSEC of 0.6057,RMSEP of 1.2901,indicating high accuracy and strong prediction ability.The eatablishment of this model provides a reference for rapid quality inspection of silicate mineral medicines.（4）The application of advanced intelligent SVM algorithm in traditional NIRS analysis.In establishing the NIRS qualitative and quantitative models of the 8 silicate mineral medicines,the intelligent SVM algorithm has prominent algorithm advantage due to its strong nonlinear fitting ability.This is a new tentetive application of the intelligent algorithm in identification of traditional Chinese medicine,and demonstrates the feasibility of NIRS and SVM in qualitative and quantitative analysis of the eight silicate mineral medicines.In conclusion,the basic data for accurate XRD-based identification of the sources of 8 silicate mineral medicine were determined by combining conventional identification methods.On this basis,NIRS qualitative method and rapid quantitative analysis method for 8 silicate mineral medicine were established based on SVM artificial intelligence algorithm,The proposed method has high analysis accuracy and overcomes the problem that silicate mineral medicine can only be decomposed at high temperature.This study provides reference for analysis of mineral medicine.