Towards Multi-Fabric Garment Detection

Textile recycling is crucial for environmental sustainability, but accurately sorting post-consumer garments by material composition remains a challenge, especially for multifabric garments. Current methods struggle to identify and separate regions with materials distinct from the main body fabric, such as prints, collars, or waistbands and rely on manual pre-sorting. This paper proposes a system that combines near-infrared (NIR) spectroscopy and camera-based computer vision for automated material detection in textiles. The proposed method leverages a pre-trained Mask R-CNN model to perform fine-grained garment segmentation. Each segment represents a region within a garment with a potentially unique material composition. NIR spectroscopy measurements are then taken on strategically chosen points within each segment for actual material characterization. Our prototype system demonstrates the potential for this combined approach to improve the accuracy and efficiency of textile sorting for recycling applications.