SBIR Phase II: Automated Textile Sorting Systems for Precision Resale and Closed-Loop Material Flow

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase II project is to improve how discarded textiles are sorted, enabling more clothing to be reused, resold, or used as secondary source of raw materials. Today, most textile sorting is done manually, which limits how much material can be processed and leads to valuable garments being discarded. This project will develop artificial intelligence (AI) tools that help identify the material composition and resale potential of textiles more quickly and accurately. By increasing the efficiency of textile sorting, this technology can increase the use of such materials, and support the growth of the circular economy. It also has the potential to create higher-quality jobs in sorting facilities by shifting work from manual inspection to technology-assisted operations. This project aligns with national priorities around resource efficiency and domestic manufacturing, and advances economic outcomes. The project addresses a major challenge of developing a scalable, AI-driven system for textile classification that can accurately predict both resale and potential for being secondary source of raw materials under real-world conditions. The primary innovation lies in combining hyperspectral and Red-Green-Blue (RGB) imaging with multimodal machine learning to enable automated sorting decisions across heterogeneous textile streams, a task that is difficult to replicate due to the need for large, high-quality, and domain-specific datasets. The scope of the project is to develop an integrated AI model that performs brand identification, textile quality assessment, and material identification—three key inputs required to route a garment to resale or other uses. The central technical challenge is detecting contaminants and defects that are often present in small quantities; for example, the presence of ~3% elastane can render a textile unacceptable for specific applications, while localized defects such as pilling can significantly reduce resale value. The intellectual contribution includes the development of data-centric AI approaches for weak-signal detection in textiles, including synthetic data generation, ensemble modeling, and multimodal integration of spectral and visual data. The project will also contribute to methods for defining and operationalizing garment quality through structured defect detection. The methodology involves large-scale dataset collection from industry partners, model training and validation across diverse textile samples, and iterative testing in operational settings. The system will be evaluated based on classification accuracy, robustness under varying conditions, and its impact on sorting throughput and decision-making. The outcome will be a deployable technology capable of improving textile sorting efficiency at industrial scale. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.