Study On The Detection Method Of Textile Fiber Composition Based On Near Infrared Spectroscopy

The textile industry has an important position in China's economic industrial structure.While promoting China's economic development,it also has an important impact on the national quality of life.With the rapid development of the textile industry,the demand for textile inspection has also increased year by year.Among them,the detection of textile fiber components is a key item in fabric testing.However,the current main detection methods are difficult to meet the growing demand for textile component testing.An accurate and high-efficiency fabric component detection method,however,currently commonly used detection methods are inefficient and difficult to operate,and it is difficult to meet the current huge detection requirements.Near-infrared spectroscopy analysis technology has high detection accuracy,fast detection speed and no damage to samples.It is currently widely used in the field of chemical analysis.This paper takes cotton-polyester two-component textiles as the research object,and conducts related research on the quantitative analysis of fabric composition by near infrared spectroscopy.The main contents include:1.Researched the principles and characteristics of near infrared spectroscopy analysis technology,understood the current research situation of near infrared technology and its application in the field of fabrics,combined with its technical principles to verify the feasibility of near infrared analysis of fabric fiber components,and provided for this research Theoretical basis.2.Designed a near infrared data collection experiment for fabric samples.After measuring the spectral data of all samples,the relevant pre-processing methods were used to process the spectrum,and the processed spectral data set was used to establish a partial least squares correction model.Preliminary prediction of the composition of fabric cotton.3.Researched two kinds of feature optimization algorithms,namely the principle of continuous projection algorithm and particle swarm algorithm,and used two optimization methods to extract feature wavelengths of spectral data,in which the number of feature wavelengths extracted by continuous projection algorithm is 17,The particle swarm algorithm is 48,and the feature extraction effect is obvious;using two characteristic wavelengths for partial least squares regression(SPA-PLS and PSO-PLS),and comparing the results of the full-spectrum model,it is found that both algorithms have very good effects on the model good lifting effect.4.Aiming at the problem that the particle swarm optimization algorithm extracts many features,the variable inertia weights are introduced for further research,three weight change functions are proposed and the fabric spectral data are also optimized.Finally,the results of several methods are found to be similar.It is believed that the particle swarm algorithm achieves feature extraction without falling into the local optimum;Combine the continuous projection algorithm and particle swarm algorithm to extract features from the spectral data.Although the final number of feature variables decreases,the performance of the prediction model decreases.