Content measurement of textile mixture by near infrared spectroscopy based on BP neural network

Abstract

An approach using near infrared spectroscopy (NIR) combined with back propagation (BP) neural network for the accurate measurement of fiber contents of textile mixture was put forward. 56 samples with different cotton and terylene contents were prepared, which were divided into the calibration set, validation set and prediction set respectively and their near infrared spectra were obtained. The wavelet transform (WT) was utilized for the spectra data compression. Multivariable linear regression (MLR) model based on the Lambert - Beer's law and BP neural network model based on WT were developed. It indicates that the prediction accuracy of WT-ca3-BP network model is 2% for calibration samples and 4% for validation samples, which is much higher than the MLR model and is suitable for the prediction of unknown samples. On the basis of not changing the structure of the WT-ca3-BP network model, calibration and validation samples were utilized fully to be re-set to new calibration samples, which upgraded this model. The upgraded WT-ca3-BP network model was applied to predict unknown samples. Experimental results show that this approach by Near Infrared Spectroscopy based on BP neural network can be used to quantitative analysis for textile fiber.