Environmental pollution leading to global warming is no longer a distant threat; it is knocking on our door, demanding immediate action. Among the major contributors to this crisis is plastic pollution. While there is increasing encouragement, including through popularisation, education, and policy measures—towards the use of biodegradable and recyclable materials, current waste management systems are unable to support these efforts. Current technology is insufficient for sorting processes to be optimally effective, resulting in continued diversion of valuable plastics into landfills or incinerators, causing irreversible harm to the natural environment and undermining the principles of a circular economy, in which materials should be kept in circulation, never becoming waste.

In the UK’s current waste management infrastructure, pristine plastics free from contaminants can be detected and separated with relative ease using conventional technologies like near infrared (NIR). However, plastics contaminated with food waste, which are commonly found amongst disposed packaging, present a significant challenge. Once contaminated, it becomes difficult to accurately identify the type of plastic, an important consideration in making these materials less likely to be recycled or composted. As a result, they are typically sent to landfills or incineration facilities, where they contribute to environmental degradation, instead of being recovered or repurposed.

Breaking the barrier with hyperspectral imaging to revolutionize plastic waste sorting

A combination of hyperspectral imaging (HSI) and machine learning offers a transformative solution to overcome the existing barriers to recycling food-contaminated plastics. Unlike NIR methods in current use, HSI delivers highly detailed spectral data, along with the capture of spatial information. Combining these capabilities enable accurate identification and classification of plastics even when contamination is present, making HSI a powerful tool for revolutionizing automated plastic waste sorting.

In this study, a machine learning model was developed to identify real-world plastic types that were heavily contaminated with food waste. The accuracy of the model, at 99%, was impressive, but an additional challenge considered was the physical features of the plastic itself, for example, how dark it appears to the imaging technology, taken level together with the degree of contamination. Our analysis showed these factors are vital to account for, as they had a significant impact on the performance of our machine-learning model: there was a slight decrease in accuracy when the model attempted to identify very dark, and highly contaminated packaging.

The implications of our study are profound. Hyperspectral imaging is a key innovation if a true circular economy in plastics is to be achieved, as a dramatic increase results from the accurate, high-throughput sorting, recycling and composting rates our system offers. If integrated into waste management facilities, recyclable plastics could be unambiguously directed to recycling plants, with biodegradable plastics sent to composting systems. In this way, valuable materials would be recovered, and organic waste would be transformed into compost, instead of being lost to landfills, as is the case at the present time.

Using hyperspectral imaging and machine learning to identify food contaminated compostable and recyclable plastics by Nutcha Taneepanichskul (University College London), Helen C Hailes (University College London) and Mark Miodownik (University College London) is published in UCL Open Environment, volume 6.

UCLOE food-contaminated plastics hyperspectral imaging (HSI) recycling composting machine learning automatic sorting UCLOE blog