Application Of Correlation Analysis Technology On Fast Inspecting Of Textile Fibre By Near Infrared Spectroscopy

This study is mainly about the application of correlation analysis technology on fast inspecting of textile fibre by Near Infrared spectroscopy analysis. After confirmed the reasonable method to mix the sample, the best instrumental parameters and compatible pretreat method for data, the correlation curves of the original spectrum in different state were discussed. To prove the effect of correlation analysis technology in quantitative analysis of textile fibre, 5 group two-mixed samples, 4 three-mixed samples and complex background data set were discussed respectively. All of the original spectrum after the first derivative, Savitzky-Golay smoothing, normalizing and correlation analysis, were built calibration models using PLS regression analysis, which prediction results were compared with the models not processed by the correlation analysis. After correlation analysis, the prediction correlation coefficient are all above 0.985, prediction average error are less than 2.300 and the largest absoluteness error are within 5, that accord with the measure error of the textile. The purification and strengthen ability for available signal of correlation analysis technology was proved.Base on the practical instance of the difficultness to find the pure simple for detection, the pure simple was replaced by the two-mixed "simple", and its effect was proved by the calibration models of three-mixed samples. That established the base of the study on multi-mixed "simple", and would popularize the application of correlation analysis. The result indicated that, as the pure "simple", the two-mixed "simple" could strengthen the signal and advance the precision and solidity of model.At last, the first derivative and Savitzky-Golay smoothing method were replaced by wavelet decomposition and reconstruction method. For the wool-cotton-silk mixed sample, after wavelet decomposition and reconstruction and correlation analysis, the prediction precision of the calibration model was improved. The prediction average error was reduced from 1.983 to 1.842, and the largest absoluteness error was reduced from 4.366 to 4.190. The result showed that the combination of wavelet decomposition and reconstruction and correlation analysis method could enhance the effect of correlation analysis for signal purification and strengthen, and it could be considered as a good method combined with correlation analysis technology in the future.