Mountains of discards reshape the landscape outside Alto Hospicio, Chile. Image courtesy of National Geographic (ref 7)

Using Hyperspectral Imaging to Sort Textiles

92M Tons
Textiles into landfills annually (global)10%
Textile production contribution of global carbon emissions

Challenge

According to the U.S. Environmental Protection Agency, 17 million tons of textile were generated in the US in 2018. Two and a half million tons were recycled, three million tons were combusted for energy, and eleven million tons were sent to the landfill¹. Thus, about 65% of waste textiles end up in landfills.

Textile waste is increasingly recognized as a significant environmental problem that currently does not have clear solutions. In 2020, the EU adopted a new circular economy plan². Part of this plan focuses on textiles, with a goal of all textile products placed on the EU market being long-lived and recyclable by 2030. The commission is also proposing targets for recycling of textile waste and supports a legislative target for textile recycling and reuse³. It is a multi-pronged and incremental approach that also includes regulating the manufacturing industry⁴. Recycling will be a key component of reducing textile-related environmental impacts and it requires dependable sorting technology.

Recycling Solution

Our internal feasibility study shows that hyperspectral imaging provides sufficient contrast to sort most textiles based on their spectral signature in the NIR spectral range (from 900 to 1700 nm), mostly irrespective of their visible color. Therefore, hyperspectral imaging has the potential to be a crucial technology for sorting textiles for recycling.

Application Overview

We scanned two dozen textile samples of natural, synthetic, and blended fibers. We then refined several data pre-processing steps and chose classification algorithms to optimize the system’s ability to distinguish between various textiles.

After model training, we successfully classified various chopped textile fibers and a few full clothing items in real-time on our experimental conveyor station. See the video below.

Equipment Used

We used Resonon’s Real-time Vision System (RVS) software and a Pika IR to scan textile samples. Scanning was completed with our SpectralSight™ System using two 4-fixture halogen lighting units.

Spectra of various textiles scanned by Resonon

Results

Synthetic fibers are the easiest to distinguish due to their well-defined and unique chemical bonds, which lead to characteristic spectral signatures that are readily classified by machine learning algorithms.

Natural fibers such as cotton and linen are made of similar cellulose building blocks. Their spectral signatures are distinct from the synthetic fibers but more similar with respect to each other.

There are several hybrid textiles, such as viscose and rayon, that are chemically processed fibers from natural sources. Since they are made of natural fibers (wood or other plant-based fibers) their spectral signatures are closer to those of cotton and linen.

Wool has a unique spectral signature with a broad dip in reflectance around 1000 nm that was not present in any other fibers. As a result, wool is easily sorted from other fabrics.

The video below shows RVS classifying chopped textile fibers and a few full clothing items in real-time.

Despite having never seen the clothing items prior to this test, the system unambiguously classifies Polyester shorts and wool socks, and the polyester band on the wool socks.

Some textile datacubes are available on downloads.resonon.com. You can download a free copy of Spectronon™ and the datacubes there.

For this case study, securing an adequate training sample set was more time-consuming than the study itself. It took less than 30 minutes to scan samples, train a classification model, and start classifying textiles with SpectralSight™.

To further advance this preliminary textile sorting system, the next step would be to train the RVS software with more samples, representative of the wide variety of existing textiles including blends like cotton/polyester and wool/acrylic.

Nonetheless, based on the success of this study, it is safe to conclude that textile sorting devices that utilize hyperspectral imaging will be robust, accurate, and reliable.

Contact Resonon to discuss your application and how we can help you solve challenges with hyperspectral imaging.