Quantitative Prediction of Cotton and Wool Mixture Materials by BPNeural Network and NIR Spectrometry

摘要

An approach of using near infrared spectroscopy combined with BP neural network method was investigated for the prediction of fibre contents of textile mixture materials. The near infrared spectra of 56 textile mixture samples with different cotton and wool contents were obtained, in which 41 samples were used for the calibration set, 10 samples were used for the validation set, while 5 for the prediction set. The wavelet transform (WT) was utilized for the spectra data compression, which combined with BP neural network (BP) was specially introduced. According to the standards of absolute error (AE), mean absolute error (MAE) and root mean square error (RMSE), a calibration model of WT-ca3-BP (41-17-2) was achieved for prediction of fibre contents of textile mixture materials. The calibration set was in combination with validation set as a new calibration set, an upgraded WT-ca3-BP (51-17-2) model appeared, its mean absolute error (MAE) was less than 0.41%, root mean square error (RMSE) was less than 0.54% and a satisfying prediction precision was achieved for unknown samples. The results indicated that near infrared spectroscopy could be successfully applied for prediction of fibre contents of textile mixture materials and upgraded WT-ca3-BP model could achieve a best prediction results.