Researchers from Jinan University and Guangzhou Customs Technology Center have developed a cost-effective UV-vis spectroscopy and machine learning method to accurately identify recycled PET content as low as 10%, advancing sustainable packaging and circular economy efforts.
A recent study explored how to identify recycled content in polyethylene terephthalate (PET) samples. This study, which was published in the journal Food Packaging and Shelf Life, was led by Qi-Zhi Su and Qin-Bao Lin from Jinan University and Guangzhou Customs Technology Center (1). In their study, the researchers demonstrated how by combining ultraviolet-visible (UV-vis) spectroscopy with advanced machine learning algorithms, they could determine virgin from recycled PET. Their work demonstrated the utility of their classification model to identify recycled content in PET samples at concentrations as low as 10% (1).
Currently, plastic waste reduction is a top priority. Because of anthropogenic activities, humans are disposing of more and more plastic waste each year. In 2025, a reported 400 million tons of plastic are produced each year, and only 9% are properly recycled (2). Approximately 11 million tons of plastic enter the oceans each year (2). On top of increased human activity, environmental regulations have become stricter. Recycled PET (rPET) is central to these initiatives, especially in food packaging, where safety, quality, and traceability are important (1). However, the widespread application of rPET in China has been constrained by the absence of cost-effective, reliable methods for determining recycled content (1). Su and Lin’s research directly addresses this gap with a new approach that is designed to improve plastic recycling efforts.
PET plastic bottle recycling bin | Image Credit: © Yeongsik Im - stock.adobe.com
In their study, the researchers used UV-vis spectroscopy, which is a relatively inexpensive and widely available technique, to examine both virgin and recycled PET pellets. From the information gleaned from the spectroscopic data, the research team revealed distinct patterns between virgin and recycled materials, which were attributed to differences in processing histories and the accumulation of oligomers in rPET (1). These subtle but consistent spectral variations laid the foundation for model development.
Out of all the data preprocessing strategies that were tested in the study, baseline removal (RMBL) emerged as the most effective. RMBL helped improve the clarity and interpretability of the spectral data. Subsequently, the team employed principal component analysis (PCA) for dimensionality reduction and paired it with the Random Forest (RF) algorithm (1). This “RMBL+PCA+RF” model demonstrated the highest classification accuracy in distinguishing 100% virgin from 100% recycled PET in binary classification scenarios (1).
Apart from testing the classification accuracy, Su and Lin developed a multi-classification model capable of detecting recycled content at varying proportions, down to just 10% (1). This level of sensitivity is a marked improvement over existing techniques, which often fail to identify low-level rPET content accurately (1).
A classification model that can reliably detect small amounts of recycled material not only enhances compliance monitoring, but another positive externality is that it can stabilize pricing in the recycled plastics market by preventing mislabeling and fraudulent mixing (1).
In their study, the model developed by the researchers was able to maintain its high accuracy across diverse material inputs, underscoring its potential utility in global supply chains (1). However, the authors caution that certain variables, particularly color, may affect model performance (1). Because the visible light region plays a significant role in the spectral analysis, colored PET samples could skew the results. Although sample decolorization is a possible workaround, the team notes that further research is needed to optimize this step and develop a fully quantitative model capable of not only identifying but precisely measuring recycled content (1).
This study shows how UV-vis spectroscopy with ML algorithms can help lead to the broader adoption of recycled materials in sensitive applications such as food packaging. Given the focus on environmental sustainability, the information from this study can be used to tackle real-world applications (1). It is expected that future studies will likely focus on refining quantification models and addressing color variability, improving applicable rPET detection systems so that they can be more widely implemented and effective.
Hyperspectral Imaging for Walnut Quality Assessment and Shelf-Life Classification
June 12th 2025Researchers from Hebei University and Hebei University of Engineering have developed a hyperspectral imaging method combined with data fusion and machine learning to accurately and non-destructively assess walnut quality and classify storage periods.
AI-Powered Near-Infrared Imaging Remotely Identifies Explosives
June 11th 2025Chinese researchers have developed a powerful new method using near-infrared (NIR) hyperspectral imaging combined with a convolutional neural network (CNN) to identify hazardous explosive materials, like trinitrotoluene (TNT) and ammonium nitrate, from a distance, even when concealed by clothing or packaging.
New NIR/Raman Remote Imaging Reveals Hidden Salt Damage in Historic Fort
June 10th 2025Researchers have developed an analytical method combining remote near-infrared and Raman spectroscopy with machine learning to noninvasively map moisture and salt damage in historic buildings, offering critical insight into ongoing structural deterioration.
Harnessing Near-Infrared Spectroscopy and Machine Learning to Detect Microplastics in Chicken Feed
June 5th 2025Researchers from Tianjin Agricultural University, Nankai University, and Zhejiang A&F University have developed a highly accurate method using near-infrared spectroscopy and machine learning to rapidly detect and classify microplastics in chicken feed.
