Category identification of textile fibers based on near-infrared spectroscopy combined with data description algorithms

Abstract

Cashmere is a kind of luxury fiber produced by goats and has high economic value. The temptation of huge profits makes it a common phenomenon to fake cashmere with cheap materials. There is increasing demand to develop simple methods for distinguishing cashmere with other animal fibers. The feasibility of combining near-infrared (NIR) spectroscopy and three kind of data descriptions, i.e., support vector data description(SVDD), k-nearest neighbor data description (KNNDD) and GAUSS methods, for this goal is explored. The Relieff algorithm is used for variable selection and principal component analysis (PCA) is used as an exploratory tool and feature extraction. A total of 395 samples belonging to four categories were collected for the experiment. The number of samples used for model construction are 69, 71, 61 and 50 for A, B, C and D as the target class, respectively. Based on the selected 67 variables and only two principal components (PCs), three types of data descriptions are obtained. The SVDD model exhibits the most flexible and tightest boundary and also achieves 100% sensitivity on the independent test set. It indicates that NIR combined with SVDD and Relieff is feasible for category identification of different animal fibers.