Abstract:

In view of the current need of rapid, nondestructive and on-site? an identification of textile composition, an identification method using hyperspectral imaging system combined with chemometry was proposed. The hyperspectral images of 10 categories of commonly used textile were captured and spectral data were extracted after hyperspectral image calibration. On the basis of comparative analysis, comparative analysis of data pretreatment and sample selection methods, a partial least squares discriminant analysis model was established to identify textile, and a technical route of identifying textile with hyperspectral imaging technique was put forward. The results show that the first derivative pretreatment eliminates the baseline drifts caused by textile processing and test conditions, thus improving the generalization performance of the identification model and reducing the representativeness requirements of training samples. Taking advantage of the built identification model, textile suffered from various processings can be identified with the accuracy rate of 96.78%, which confirms the feasibility of hyperspectral imaging technology in textile components qualitative identification.

Key words: hyperspectral imaging, textile, chemical composition identification, partial least squares discriminant analysis