Fuzzy pattern recognition and classification of animal fibers

Abstract

Several techniques, including chemical and physical approaches, have been previously developed to differentiate between animal fibers. Since all animal fibers are comprised of essentially the same keratin, they cannot be effectively distinguished by existing physical or chemical technique. A fuzzy neural pattern recognition system is developed to classify two typical animal fibers: mohair and merino. Two multilayer networks are used, with the unsupervised network being used for automatic feature extraction and the supervised network serving as the classifier based on the information extracted from unsupervised network. It is found that this hybrid network can accurately classify the two fibers and the accuracy improves with the increase in the features being extracted from the unsupervised network.