Abstract:

Aiming at problems that conventional textile testing is influenced by the senses and mental status of testers, the detection results of the same sample are different from different testers, and fiber is damaged by most conventional detection methods, a method for identifying and classifying eight kinds of single component textile materials were designed based on hyperspectral imaging technology.  After the pretreatment of the hyperspectral data of the textile, the characteristic wavelengths of the various textile materials were extracted by continuous projection algorithm. The original 288 wavelength data in 920-2500 nm were compressed to 5-7(the data is compressed to 1.74% -2.43%). The least squares support vector machines are used to establish two classifiers for each kind of textile, and the extracted characteristic wavelengths were imported into the corresponding classifiers. Finally, the test samples were classified and identified. The experimental results show that 640 experimental samples are identifide by two kinds of classifiers. Hyperspectral imaging technology can be applied to the material identification of cotton,polyester, polyethylene, wool, polyvinyl chloride, nylon, linen and silk.

Key words: textile, single component identification, hyperspectral imaging, spectral analysis, continuous projection algorithm, least squares support vector machine