R. Kumar, S. Singh, R. Rao, L. Pal
North Carolina State University,
United States
Keywords: hyperspectral imaging, characterization, textile waste, machine learning, sustainability
Summary:
Efficient textile recycling is hindered by the absence of automated, fiber-specific sorting technologies, resulting in mixed and contaminated waste streams. This study demonstrates the potential of hyperspectral imaging (HSI) combined with machine learning (ML) for the automated identification of fabric types and contaminant regions in textile waste. HSI was employed as a nondestructive method to capture the spectral signatures of eight pure fabrics and corresponding apparel, as well as samples with simulated contaminants. Endmember extraction enabled differentiation of unique chemical compositions, while principal component analysis (PCA) revealed distinct spectral clustering among fabric types. The resulting spectral dataset was used to train ML classifiers to predict fabric composition and contamination in unknown samples. The model was tested on clean apparel items and post-consumer textile waste to predict the fabric types and localized contaminants on the sample. This can further the automated fiber sorting and closed-loop textile recycling within a circular economy framework.