Using Neural Networks and NIR Spectrophotometry to Identify Fibers

Abstract

A qualitative nondestructive technique for fiber identification was developed using near infrared (NIR) spectroscopy. A neural network was trained to identify 17 different fiber types using the NIR absorbance spectra from a library of 390 samples. The neural network model was verified by testing untrained samples. It was not only able to identify single fibers, but was also able to correctly identify blends of fibers from fabrics

Near-Infrared Spectroscopy for Anticounterfeiting Innovative Fibers

“…Ghosh and Roy [3] used homologs of cotton to develop a calibration equation for monitoring the sugar content in cotton; Gosh and Rodgers [4] and Tincher et al [5] investigated the heatset temperature of nylon carpet and its heat history using NIR spectroscopy. By using advanced diagnostic statistics and computer programs, Richard et al [6] and Jasper and Kovacs [7] demonstrated the qualitative classification of various natural and synthetic fibers to reveal subtle differences among NIR spectra in the set of samples. Sohn et al [8] and Ruckebusch et al [9] used NIR spectroscopy in quantitative analyses of linen/cotton and cotton/polyester blends, respectively.…”

Artificial Neural Networks and Their Applications in the Engineering of Fabrics

“…In the case of synthetic fibres, the ANNs have supported the identification of the production control parameters (Allan et al, 2001) and the prediction of the properties of the melt spun fibers (Kuo, 2004). ANNs have been used in conjunction with NIR spectroscopy for the identification of the textile fibres (Jasper & Kovacs 1994). A system for the optimization of the yarn production based on the blend characteristics and the process parameters has been developed based also on the use of ANNs (Sette & van Langenhove, 2002).…”

Non-Supervised Neural Categorisation of near Infrared Spectra. Application to Pure Compounds

“…Second, all substances are correctly identified and the total number of occupied neurons remains closer to the number of classes, which implies a good clustering. For higher size values (8,9,10), although all the substances are correctly identified, every spectrum tends to be identified as an independent class; that is, samples of the same substance are recognised by more than one neuron. This effect can be explained as a specialisation process.…”

“…Only in some very recent papers, can the use of this technique for qualitative analysis be found. [7][8][9][10] The drawback of back-propagation networks is that they require supervised training. This means that the network needs to know in advance the composition or identity of the samples in the training data set.…”