Unsupervised grouping of industrial textile dyes using K-means algorithm and optical fibre spectroscopy

Abstract

A method for the unsupervised clustering of optically thick textile dyes based on their spectral properties is demonstrated in this paper. The system utilizes optical fibre sensor techniques in the Ultraviolet-Visible-Near Infrared (UV-Vis-NIR) to evaluate the absorption spectrum and thus the colour of textile dyes. A multivariate method is first applied to calculate the optimum dilution factor needed to reduce the high absorbance of the dye samples. Then, the grouping algorithm used combines Principal Component Analysis (PCA), for data compression, and K-means for unsupervised clustering of the different dyes. The feasibility of the proposed method for textile applications is also discussed in the paper.