Study On Determination Of Fiber Content Of Cotton Poly Blended Fabric By Spectroscopic Method

As a daily necessity,the quality of fabric is closely related to people’s health.The fiber content of fabric directly affects its quality.In recent years,the market sampling results are worrying,the fabric fiber composition or content does not conform to the label,adulteration and other problems are prominent,seriously disturb the market order,so it is of great significance to test the fabric fiber content.Although the traditional chemical dissolution method is accurate,it is time-consuming,complicated and costly.With the endless variety of fabrics,the number of submitted for inspection is increasing year by year.It is urgent to research and develop a simple and rapid detection technology.near infrared(NIR)spectroscopy has been applied to the rapid detection of fabrics because of its advantages of fast,high efficiency and non-destructive.Multivariate correction partial least squares(PLS)is the most commonly used method for quantitative analysis of blended fabrics by near infrared spectroscopy.However,it is necessary to collect a large number of samples to establish a sound and robust PLS model with good adaptability.The wide variety of blended fabrics increases the difficulty and cost of PLS modeling for blended fabrics,which hinders the large-scale application of this method.In order to solve this problem,this paper takes the cotton poly blend fabric as the research object,uses near infrared spectroscopy combined with spectral extraction method to quantitatively analyze the content of cotton fiber and polyester fiber in the cotton poly blend fabric,and establishes a small sample modeling method for the blended fabric.The main research contents and conclusions are as follows:(1)The samples of cotton poly blends with different cotton fiber contents were collected,and the spectra were collected by NIR spectrometer.The quantitative modeling effect of PLS on the cotton fiber and polyester fiber contents in small samples in blends was studied.151,30 and 9 correction sets were selected to establish quantitative models of cotton and polyester fiber content.Rp values of PLS model of cotton fiber were 0.996,0.989 and 0.989,respectively.The standard deviation of SEP was 1.520,2.486 and 2.997,respectively.Rp values of PLS model of polyester fiber were 0.996,0.989 and0.989,respectively.SEP was 1.521,2.487 and 2.997,respectively.The comparison of modeling results of different numbers of calibration set samples shows that the model prediction performance decreases with the reduction of modeling sample size.Therefore,a large number of samples need to be collected to establish a sound,robust and adaptable blended PLS model.(2)In view of the large sample size required by PLS modeling,the oblique projection method is used to establish a small sample quantitative model for quantitative analysis.In this method,the spectral background signals of pure cotton samples and pure polyester samples were established,and appropriate samples of pure cotton and pure polyester samples were selected as the spectral signals of pure components to calculate the oblique projection operator.Then,the spectral signals of pure components of cotton and polyester in the blended spectrum were separated by the oblique projection operator,and the calibration curve was established and predicted with the corresponding fiber content.Taking 30 cotton polyblends as modeling samples,the prediction results of cotton fiber model based on the existing oblique projection algorithm R~2=0.8167,SEP=7.249;The polyester fiber model predicted R~2=0.9539,SEP=3.654.The prediction performance of the model is poor.In this paper,a model optimization method for background signal improvement and feature band selection is proposed.The prediction results of the optimized cotton fiber model R~2=0.9765,SEP=2.603;Polyester fiber model predicted R~2=0.9806,SEP=2.384,R~2 significantly increased,SEP significantly decreased.The results show that the prediction accuracy of cotton and polyester fiber models can be improved successfully by the oblique projection optimization algorithm.The prediction accuracy is equivalent to the PLS prediction accuracy of 30 sample models,and the prediction accuracy of oblique projection polyester model is better than the PLS model.The calibration curves of cotton and polyester fiber content of 9 blended samples were established and predicted by the optimization algorithm.The prediction results of cotton fiber R~2=0.9811,SEP=2.553;Polyester fiber R~2=0.9839,SEP=2.403.It was found that SEP was comparable to the PLS model established by 30 samples and the optimized oblique projection model,indicating that the near infrared spectroscopy combined with the improved oblique projection method can realize the quantitative analysis of the cotton and polyester fiber content in the cotton poly blend with a few samples,reducing the collection of modeling samples,and the results are more accurate.The quantitative analysis of a variety of blended fabrics can be realized only by establishing a pure component background sample library and collecting a few blended fabric samples,thus reducing the difficulty and cost of modeling.(3)Aiming at the problem that oblique projection method also needs to collect background sample library,a new projection method(NMP)is proposed.Based on 30 and 9 cotton poly blends,the spectral components of cotton and polyester fibers in blends were extracted by NMP method and standard sample spectra,and the quantitative calibration curve and prediction were established based on characteristic peak intensity and corresponding fiber content.The R~2 values of cotton fiber were 0.9457 and 0.9534,and the SEP values were 4.017 and 4.197,respectively.The predicted R~2 values of polyester fiber were 0.9505 and 0.959,and those of SEP were 3.832 and 3.851,respectively.The results show that this method can also be used to quantitatively analyze the content of cotton and polyester fiber in cotton polyblends with fewer samples.Compared with the PLS and optimized oblique projection methods in the first two chapters,this method does not need to collect a large number of modeling samples and background sample libraries,which greatly reduces the cost and difficulty of modeling.However,the prediction accuracy of this method is still insufficient at present,and the deviation of prediction results is large,which needs to be further optimized and improved.