Recent research published in journals resources, conservation, recycling discuss how the fusion of spectroscopy and advanced machine learning (ML) can dramatically improve the efficiency and accuracy of plastic recycling systems (1). The study, led by Washington State University Tri-Cities researchers, advances sustainable materials management and reducing the environmental footprint of plastics.
World plastic production is increasing every year. In 2020, global plastic production was approximately 435 tons (2). Based on current trends, global plastic production is expected to reach 736 tons by 2040 (2). As a result, the global plastics market is experiencing steady growth of approximately mid-single digits, with an expected CAGR of 5.1% from 2025 to 2032 (3).
Currently, the majority of consumer plastics such as polyethylene terephthalate (PET), high-density polyethylene (HDPE), polyvinyl chloride (PVC), low-density polyethylene (LDPE), polypropylene (PP), and polystyrene (PS) end up in municipal waste treatment facilities and landfills, causing long-term environmental problems.
In this study, researchers sought to solve the problem of downcycling, which is the process of reusing materials to create something of low value (4). Although downcycling is better than treating plastic as pure waste, it does not lead to true material recovery. Current sorting methods often struggle with contaminated, dyed or weathered plastics, and researchers have developed a new method that can solve these problems.
Their approach involved developing an ML framework rooted in convolutional neural networks (CNNs) that leverages its strengths in analyzing high-dimensional data such as spectroscopic signals. The majority of research has focused on vibrational spectroscopy, particularly Raman scattering and attenuated total internal reflection-Fourier transform infrared (ATR-FTIR) spectroscopy (1). Raman spectra collected from real recycled samples were used to train a six-class CNN that can distinguish between the most common consumer plastics.
As a result, CNN achieved 100% classification accuracy using Raman data, demonstrating that deep learning can extract chemically meaningful features from complex spectra without the need for manual preprocessing or feature selection (1).
To increase its applicability in the real world, the researchers extended their approach to ATR-FTIR spectroscopy, which is widely used in field and industrial settings. An ATR-FTIR-based model using a similar CNN architecture achieved 95% accuracy, confirming that this approach is robust across complementary spectroscopic techniques (1). Additionally, the model maintained low computational cost, which is an important factor for deployment in automated, high-throughput recycling environments.
The study also investigated how four analytical techniques (ATR-FTIR, near-infrared reflectance) work. [NIR]laser-induced breakdown spectroscopy [LIBS]and X-ray fluorescence [XRF]) Identify and classify plastics. Results showed that ATR-FTIR, NIR, and LIBS were effective on consumer plastics, with success rates of 99%, 91%, and 97%, respectively (1). When applied to marine plastic debris, which is often more degraded and contaminated, ATR-FTIR again leads the field with a 99% success rate, while NIR, LIBS, and XRF achieved 81%, 76%, and 66%, respectively (1).
The main takeaway from this study is that the researchers used CNN to identify plastics. Although CNNs are well established in areas such as image recognition, their application to spectroscopic datasets for plastic classification is still relatively new (1). CNNs can autonomously learn relevant spectral features, eliminating the need for expert spectral interpretation or manual feature engineering. This not only increases efficiency but also improves scalability across different equipment and operating conditions.
Equally important is the model's resilience to real-world fluctuations. Researchers found that CNN successfully handled plastics containing dyes, additives, and signs of environmental degradation, factors that routinely confound traditional sorting techniques (1). According to the researchers, this robustness is an important step toward field-ready systems that work reliably at recycling facilities, material recovery facilities, and even marine debris monitoring operations (1).
“This model effectively processes dyed plastics, plastics with additives, and plastics exposed to real-world environmental conditions and degradation,” the authors write in their research paper (1), “representing a significant advance that is practical for real-world applications.”
References
- Congressman Garcia Tovar. Villarreal Blanco, Massachusetts; Primera Pedroso, Orm; et al. Identification of common types of plastics by vibrational spectroscopy techniques. resolution con. record 2026, 227108767. DOI:
10.1016/j.resconrec.2025.108767 - Dabo, M. Global demand for plastics shows no signs of slowing down. packaging gateway. Available at:
https://www.packaging-gateway.com/features/global-lastic-demand-shows-no-signs-of-slowing/ (Accessed 5 January 2026). - rePurpose Team, Plastics Market Outlook 2025: What does regulation, changing demand and UN treaties mean? reuse. Available at:
https://www.rePurpose.global/blog/ Plastics-market-2025-outlook-what-will-regulation-demand-shifts-and-a-un-treaty-mean (Accessed 5 January 2026). - Cary Company, What is Downcycling? Cary Company. Available at:
https://www.thecarycompany.com/insights/articles/what-is-down-cycling?srsltid=AfmBOoqpImDvlsFCrLfrYHkynyV8AvKXXEumnhpwoQ-pkCcOuDsGss5p (Accessed 5 January 2026).
