Data treatment method for the spectrum data of textile

摘要

The NIR region is composed of radiation with wavelengths of 700-2500nm. The analytical technology of NIR has many virtues, such as fast (one minute or less per sample), nondestructive, suitable for on-line use. So it can be applied to the textile field. But because of the interference from strongly overlapping constituents' spectra and from light scatter variations, the transformations of the diffuse spectroscopy measurements should ideally pass through two stages, response linearization and optical correction. Before being used in linear calibration model, the spectra data usually is pretreated by the different pretreatment methods. The pretreatment methods contain derivative, smoothing, normalizing, data compression and so on. These pretreatment methods resolve the overlapping peaks, remove the linear baselines and eliminate the spectral noise. Then three methods, Multiple Linear Regression (MLR), Partial Least-Squares (PLS) and Neural networks are adopted to establish a model to with the pretreated spectra data. The first two methods express a linear relationship between the spectral data and the concentration. And the third method is a nonlinear method. The validation sample set is used to validate these three established models. Depending on the comparison of the results, the best linear calibration model to estimate the unknown samples is set up.