Towards Improved Sustainability in The Textile Lifecycle with Deep Learning

Summary

The paper details a study addressing the critical sustainability challenges posed by the garment industry, specifically focusing on textile recycling and traceability to enable a circular economy. The authors propose a cost-effective solution leveraging deep learning and low-cost hardware for fabric identification and traceability. They achieve fabric classification accuracy of over 90% for 14 fabric classes using microscopic images and Convolutional Neural Networks (CNNs). They also develop a UV-visible coding system, readable by YOLOv8 object detection models, demonstrating impressive mean Average Precision (mAP) scores—over 0.98 for fresh codes and up to 0.93 after multiple wash cycles. Furthermore, a prototype reader application was introduced, showcasing the practical implementation of these methods for global scalability in fabric lifecycle management.

The significance of this research lies in its potential to drastically improve textile recycling rates and enable enforceable circular economy legislation by providing reliable fabric identification and traceability solutions. The CNN-based fabric classification method enhances sorting efficiency while the UV marker-based traceability mechanism ensures that garments can be tracked throughout their lifecycle, even after several wash cycles. The authors also stress the importance of scalability and affordability, making the proposed solutions accessible for broader implementation. The study concludes with a recognition of the need for further research, particularly in handling fabric blends and ensuring marker longevity and safety, with the promise of contributing valuable datasets for ongoing and future research in this domain.