Fabric Composition Classification Using Hyper-Spectral Imaging

This paper explores the use of Hyper-Spectral Imaging utilising multiple visual bands in the Short Wave Infrared range to analyse fabrics and identify their composition materials. This paper proposes using classical computer vision and modern machine learning techniques, in order to classify the composition of fabric materials. Using unsupervised segmentation, data from high resolution images is reduced for a supervised classifier. Segmentation and classification is compared among eight different combinations of classical and machine-learning solutions. Segmentation is done using Felzenswalb and Kmeans methods, while classification uses the Minimum Spectral Distance, Spectral angle mapping, Principal component analysis with a support vector machine and a convolutional neural network.