Research On The Construction Method Of Hyperspectral Intelligent Analysis Model For Blended Fabrics

With the development of textile international trade and consumers’ pursuit of fabric quality,China’s textile testing standards and requirements are becoming more and more strict.At present,spectral analysis technology combined with stoichiometry has been widely used to establish calibration model to analyze the composition and content of textiles.In general,the qualitative analysis model is used to identify the types of textiles,and then the corresponding quantitative model is used to predict the content,so as to achieve the purpose of detection.In order to solve the blind selection model in the textiles testing component analysis problem of prediction accuracy is poorer,and the paper has carried out based on high spectral analysis technology of the blended fabric of intelligent analysis model building method research,in order to achieve the unknown blended fabric sample to determine its composition information,directly after choosing reasonable quantitative analysis model to predict its content,That is to realize the intelligent choice of qualitative and quantitative models.Paper to cotton polyester blended fabrics as the research object,on the basis of the conventional qualitative analysis,puts forward the spectra analysis method based on LS-ICA,or components can be unknown categories of textiles for spectra analysis,by comparing with pure substance of the second derivative spectrum obtained contained component category,to choose the appropriate qualitative identification model based on component category for validation,Then choose a suitable quantitative analysis model to analyze the component content.In order to realize the construction of hyperspectral intelligent analysis model of blended fabric.The main research contents and conclusions are as follows:In order to optimize the collection parameters and ROI selection of hyperspectral images,the influence factors of textile hyperspectral images were analyzed and the method of automatic ROI selection was studied.The results show that different colors of textiles have almost no effect on the spectrum in this band,and the spectral shapes of gray image and RGB image are basically consistent.In the effective region,the different moving speed of the displacement platform only affects the display of the hyperspectral image,which makes the hyperspectral image stretch to different degrees,but has no significant influence on the spectral data.As for the number of folded layers of fabric,the spectral curves are basically the same under the experimental condition that the light cannot directly illuminate the background of the object table through the interlacing gap of the fabric.For ROI selection,and puts forward the unified extract image ROI rectangular area of the same place and the same number of pixels to the standard for subsequent spectrum data collection,through the image segmentation method,the acquisition of hyperspectral image segmentation into 2 * 3 of the same small,extract 100 pixels per center of small rectangular area as ROI selection,It reduces the contingency and randomness of traditional manual extraction and time-consuming problems.The blended fabric composition differential analysis model building problem,has carried out based on partial least squares discriminant analysis(PLSDA)polyester,cotton,polyester and viscose polyester ammonia ammonia blended fabric composition discriminant model of qualitative research,analyses the SG smoothing,SNV,MSC and CWD four consecutive pretreatment method to model the correct discriminant rate and F is worth,The results were compared with SVM and ELM models.The results show that the PLSDA model is superior to the SVM model and the ELM model,and can accurately identify the three blended fabrics,the accuracy is up to 100%,the F value is also up to 1,and the model operation speed is fast.In order to solve the problem that conventional qualitative discrimination needs to establish a discriminant model,and the parameters and characteristic variables of the model have a great influence on the results of discrimination,a qualitative discriminant model construction method based on spectral decomposition was studied,and a new model construction method based on LS-ICA combined with hyperspectral analysis of blended fabrics was proposed.In order to verify the feasibility of the model,a model of cotton-polyester blended fabric based on hyperspectral LS-ICA was established.Based on the analysis of the influence of pretreatment methods on hyperspectral data,the preferred MSC pretreatment method was used to preprocess the hyperspectral data,and then the characteristic spectral data was selected by Laplace fraction value and separated by ICA.Thus,the MODEL of LS-ICA spectral analysis of cotton-polyester blended fabric was constructed.The second derivative spectra of pure cotton and pure polyester were compared and analyzed.The second derivative spectra of pure cotton and IC1 had obvious characteristic peaks at about 1400 nm and 1920 nm.The second derivative spectra of pure polyester and IC2 have obvious characteristic peaks at about 1400 nm,1700nm and 1920 nm.The results show that the LS-ICA model can be used to analyze the spectrum of cotton-polyester blended fabrics and identify the unknown components.Finally,the blend samples are brought into the established conventional qualitative identification model for verification,and the accuracy is 100%.Aiming at the problem of constructing the analysis model of the composition content of the blended fabric after qualitative analysis and identification of the component types of the blended fabric,the research of constructing the quantitative analysis model of the blended fabric was carried out.The quantitative analysis models of polyester,cotton-polyester and viscose polyester were constructed respectively.Four methods,SG smoothing,MSC,SNV and CWD,different from concentration gradient method and SPXY,were analyzed.MCUVE,VCPA,IVISSA and IRIV were used to screen variables to improve the analytical accuracy of the quantitative model.After that,three kinds of blended fabrics of polyester,cotton-polyester and viscose polyester were predicted by the established three quantitative models.The results show that the quantitative model is generally not universal,and only when the model with the same component predicts its content,the effect is the best.That is,the quantitative model of polyester is the best,the quantitative model of cotton polyester is the best,and the quantitative model of viscose polyester is the best.Therefore,it is more necessary to construct the intelligent analysis model of blended fabric.In order to avoid randomly and blindly introducing quantitative models of different components into the samples to be tested for content prediction.For the full text,if the major categories are known,the components can be obtained through the conventional qualitative identification model analyzed in Chapter 3,and the quantitative models with the same components can be directly selected for analysis;if the major categories are unknown,go through LS-ICA performs spectral analysis of Chapter 4,determines its components(that is,known categories)by comparing the second-order derivative spectra of pure substances,directly selects and analyzes the traditional qualitative identification model of known categories for verification,and finally directly selects the same quantitative model for components.Analyze and predict its content,according to the component intelligently select the qualitative model containing the component category for verification and identification,and then according to the known component,intelligently select the same quantitative analysis model as the component to predict the content,so as to realize the model analysis of blended fabrics.Intelligent,eliminates the blindness of model selection during model analysis,solves the problem of long time and effort in selecting models,and improves analysis efficiency.