Using hyperspectral imaging and machine learning to identify food contaminated compostable and recyclable plastics
This is version 1 of this article, the published version can be found at: https://doi.org/10.14324/111.444/ucloe.3237
Abstract
With the increasing public legislation aimed at reducing plastic pollution, compostable plastics have emerged as an alternative to conventional plastics for some food packaging and food service items. However, the true value of compostable plastics can only be realised if they do not enter the environment as contaminants but instead are processed along with food waste using industrial composting facilities. Distinguishing compostable plastics from other plastics in this waste stream is an outstanding problem. Currently, near-infrared technology is widely used to identify polymers, but it falls short in distinguishing plastics contaminated with food waste. This study investigates the application of hyperspectral imaging to address this challenge, enhancing the detection and sorting of contaminated compostable plastics. By combining hyperspectral imaging with various machine learning algorithms we show it is possible to accurately identify and classify plastic packaging with food waste contamination, achieving up to 99% accuracy. The study also measures the impact of plastic features such as darkness, size and level of contamination on model performance, with darkness having the most significant impact. The developed machine learning model can detect plastic with higher levels of contamination more accurately compared to our previous study. Implementing hyperspectral imaging in waste management systems can significantly increase composting and recycling rates, and improve the quality of recycled products. This advanced approach supports the circular economy by ensuring that both compostable and recyclable plastics are effectively processed and recycled, minimising environmental impact.
Keywords: food-contaminated plastics, hyperspectral imaging (HSI), recycling, composting, machine learning, automatic sorting
Funding
- Natural Environment Research Council (grant NE/V010735/1)