Establishment And Prediction Of Near Infrared Spectral Analysis Model For Polyester-Cotton Blended Fabrics

A series of 598 polyester-cotton blend samples were studied in this article,which content was obtained by national standard method(GB/T 2910.11-2009)and their original spectrum were acquired by near-infrared(NIR)spectrometer.All of 598 cotton-polyester blend fabric samples had been divided manually into two categories,among 252 samples were used in calibration set,and the other 346 swatches were used for validation set to verify the accuracy of the model.For unknown samples qualitative analysis,In the range of 1000~2500 nm band,polyester-cotton blend fabrics NIR qualitative model was established by classification of principal component analysis Mahalanobis distance combined with other related mathematical correction method.RC was 0.999,RMSECV was 0.009,RP was 0.997,SEP was 1.793.For verify validity and practicability of the model,346 samples were chosen as an validation set,the correct recognition rate reached up to 96%.For unknown samples quantitative analysis,In the range of 1000~2500 nm band,polyester-cotton blend fabrics NIR quantitative model was established by partial least squares(PLS)method after the spectrum data pretreated by the method of chemometrics.During in progress of the model established,through different pretreatment methods were chosen and the best modeling band was selected by the distribution trend of regression coefficient,Achieve the goal of the optimized model.Main evaluation parameters of two model:(1)first derivative+S-G smoothing+Norm,which is suitable for predicting the fiber content of conventional polyester-cotton blend samples;(2)first derivative+S-G smoothing+Norm+OSC,which is suitable for the prediction of the fiber content of denim fabric with thin coating and indigo dyeing.RC were 0.996 and 0.998,RMSECV were 0.041and 0.002,RP were 0.995 and 0.997,SEP were 1.860 and 1.121,and the external validation accuracy reached up to 97%.Analyzed the reason of the occurrence of slash spectral samples and singular samples.At last,346 bound samples were predicted,the paired-samples t test(t_T=0.1440)indicated that there was no significant difference between NIR method and national standard method,analysis time less than 10 seconds.In addition,30 representative polyester-cotton blend samples were selected as the study objects,The influences of temperature,fabric color and fabric stitch structure on the original NIR spectra and the prediction accuracy of the model were discussed.The experimental results shown that the influence of temperature on the original NIR spectra of cotton and cotton fabrics with high content was significant,the dark fabrics can drift some of the baseline of the original NIR spectrum,but make no difference to model predictability.The study to fabric stitch structure indicated that only roving knit,thick coated fabric and the indigo dyed denim fabric can't be predicted accurately using conventional near infrared spectrum analysis.