Application Of Near-infrared Spectroscopy In Inspection Of Textiles

In this study,we chose 892 textiles’ samples,including the polyester,linen,cotton,silk,wool,polyester-cotton and polyester-wool,from the market as the object,and component identification and content detection in textiles were studied based on near-infrared spectroscopy(NIRS).Discriminant models of component in textiles were built based on shortwave near-infrared.In order to overcome the effects of samples’ color,sparse principal component analysis(SPCA)was proposed,which can effectively extract the information in the spectra.Discriminant models of component and detection models of content in textiles were established by using mediumwave near-infrared,the effects of smoothing methods and preprocessing methods on models were discussed,and wald-wolfowitz method was proposed to test the representative of calibration set and prediction set samples for detection models of content.Main contents and conclusions were listed as below:(1)Spectra of polyester,linen,cotton,silk,wool,polyester-cotton and polyester-wool were collected by using SupNIR-1520 and SupNIR-1100 which are special near-infrared spectrometers,and the content of polyester-cotton and polyester-wool were obtained by chemical analysis.(2)Spectral characteristics of textiles were studied.The results show that different textiles have different absorption peaks,which provides a basis for the usage of specral analysis technology in the inspection of textiles.(3)Discriminant models of polyester,linen,cotton,silk,wool,polyester-cotton and polyester-wool were built based on mediumwave near-infrared by using PLSDA method.Its result shows that the modeling effect is the best when using Savitzky-Golay piecewise smooth combined with second derivative preprocessing method.In this model,correct rate of polyester,linen,silk,wool,polyester-wool samples were 100%,and the correct rate of cotton and polyester-cotton samples were 96.55% and 96.88% respectively.(4)Detection models of content for polyester-cotton and polyester-wool were built based on mediumwave near-infrared by using PLS method.Its result shows that the modeling effect is the best when using Savitzky-Golay piecewise smooth,first derivative,and SNV combined with DT preprocessing method.The wald-wolfowitz method was proposed to test the representative of calibration set and prediction set samples,and its result shows that SPXY method is the best method to divide samples set.Finally by paired-samples T test,polyester-cotton and polyester-wool prediction set samples were tested.The results show that there is no significant difference among NIR predicted results with classical method’s results,and the models’ R2 of prediction set samples and root mean standard error of prediction(RMSEP)were 0.99,2.22% and 0.98,3.47%,respectively.(5)Discriminant models of polyester,linen,cotton,silk,wool,were built based on shortwave near-infrared.The research shows that SPCA can effectively extract spectra information,overcome the effect of model from samples’ color and improve the accuracy of the model,from 94.55% to 100%.It proves that shortwave near-infrared spectroscopy can realize the identification of components in textiles.