Reproduce “Supervised and Unsupervised Textile Classification via Near-Infrared Hyperspectral Imaging and Deep Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Maria Kainz, Johannes K. Krondorfer, Malte Jaschik, Maria Jernej, Harald Ganster

Researchers at JOANNEUM RESEARCH and Graz University of Technology applied deep learning, using both supervised convolutional neural networks and unsupervised autoencoders, to near-infrared hyperspectral imaging data for automated textile classification. The supervised model achieved perfect object-based classification across 12 textile types, demonstrating strong generalization to variations in color and structure, while the unsupervised autoencoder effectively identified pure non-target fibers, supporting advancements in textile recycling.

Key insights

  • Supervised CNNs, combined with appropriate preprocessing, exhibit robust generalization capabilities, achieving perfect object-based classification even when textile samples vary significantly in color and structural properties.
  • Unsupervised autoencoders can effectively detect pure non-target textile fibers (anomalies) with high accuracy, offering a viable solution for initial sorting steps when comprehensive labeled data is unavailable.
  • The effectiveness of unsupervised autoencoders is reduced when classifying textile blends or structurally diverse target textiles, highlighting a limitation for fine-grained classification in complex waste streams.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.