An automated textile sorting system using artificial intelligence and industrial image processing is being developed to improve the efficiency of textile recycling. The system, known as DETEX, is designed to classify used garments automatically and support mechanical recycling processes by identifying clothing types and material properties.
The technology is being developed at the Recycling Atelier of the Augsburg Institute for Textile Technology (ITA), a model factory focused on mechanical textile recycling processes. The initiative addresses a key challenge in textile waste management: the limited share of used textiles that are recycled into new materials.
Germany generates around 1.4 million tonnes of used textiles each year. Of that volume, roughly 200,000 tonnes are manually sorted and assigned to recycling streams, while the majority is either thermally recycled or exported. Manual sorting remains the dominant method despite the growing volume and diversity of materials, which includes blended fabrics and low-quality garments associated with fast fashion production.
Efficient sorting is a prerequisite for mechanical textile recycling. Before garments can be processed into new materials, they are shredded, stripped of components such as buttons or zippers, and separated into fibres. Maintaining fibre length during shredding is essential for producing higher-quality recycled material. Differences in fabric structure and areal density also influence processing requirements, making accurate material classification necessary before recycling.
The DETEX system applies AI-based image recognition to automate this classification process. Two high-resolution industrial cameras capture images of garments as they move along a conveyor belt. Neural networks analyze the images and classify items based on patterns and structural features identified during model training.
The imaging hardware used in the system includes uEye XC cameras produced by IDS Imaging Development Systems. The cameras capture detailed images of each garment, which are processed by several neural network models designed for different tasks, including object detection, garment classification and material identification.
During operation, an object detection model analyzes images from a camera mounted above the conveyor belt to identify the garment type, such as trousers, dresses or T-shirts. A second camera positioned approximately five centimetres above the textile captures additional images that focus on fabric characteristics and features such as stains or attached components.
The relevant image sections are then extracted and processed by another AI model that identifies the material structure, distinguishing for example between woven and knitted fabrics. The system displays the classification results on a screen for monitoring and downstream processing.
Training the AI models required a large dataset of manually labelled images. Each clothing category was represented by at least 3,000 sample images, including photographs of complete garments as well as close-up images of fabric surfaces. The labelled data enabled the neural networks to learn visual patterns associated with specific garment types and material structures.
The system uses the IDS peak software development kit to integrate the cameras into the image processing workflow. The software provides programming interfaces and tools for operating and controlling the cameras and enables image data to be transferred via the USB3 Vision interface for further analysis.
Researchers are continuing to expand the DETEX concept into a more comprehensive sorting platform. The current conveyor-belt configuration is expected to evolve into a modular system combining mechanical handling and robotic components. One proposed development involves a free-fall mechanism that would allow textiles to be captured from multiple angles during sorting, enabling 360-degree visual analysis.
Additional robotic grippers could further inspect garments by capturing images from both sides, allowing the system to collect more detailed information about material characteristics. The expanded approach is intended to improve classification accuracy and support more precise routing of textiles into recycling or reuse pathways.
The Recycling Atelier at the Augsburg Institute for Textile Technology operates as a research and demonstration facility for circular textile processing. The institute’s work focuses on developing technologies that support the reuse and recycling of textile materials through automated and data-driven systems.
