Don’t Give Me Any of That Trash

An artificially intelligent algorithm with superhuman senses helps recyclers sort compostable and traditional plastics.

Nick Bild
10 days agoMachine Learning & AI

Now more than ever, compostable plastics are being explored as a potential solution to the global plastic waste problem. These materials are designed to break down into organic matter when they are discarded, reducing the amount of plastic that ends up in landfills and oceans.Compared to traditional plastics, compostable plastics have several advantages that make them a promising alternative.

Traditional plastics can take hundreds of years to decompose, while compostable plastics break down in a matter of months, leaving behind no harmful residue. This means that compostable plastics can be safely disposed of in composting facilities or even in backyard compost bins, where they will break down naturally and become part of the soil.

Compostable plastics are also more sustainable than traditional plastics because they are made from renewable resources such as cornstarch, sugarcane, and potato starch. Unlike fossil fuels, these resources are renewable and can be replenished over time, making compostable plastics a more environmentally friendly option.

According to a report by the European Bioplastics Association, the global production capacity for compostable plastics is expected to grow from 1.7 million tons in 2020 to 2.7 million tons by 2025. The same report estimates that compostable plastics will account for 5% of the global plastic market by 2025.

But counterintuitively, this rise in the production of compostable plastics can actually have some detrimental impacts on the environment. Because they are virtually indistinguishable from traditional plastics visually, they create havoc for recycling facilities. These eco-friendly plastics are considered a contaminant by recyclers, as they reduce the efficiency of their operations, and reduce the value of their efforts. Moreover, confusing the plastics can also lead to traditional plastics winding up in compost heaps.

A trio of researchers at University College London recognized that we need a better, more accurate, and far more efficient means of identifying these plastics to assist us in sorting them correctly. Since this task has already proven itself too challenging for humans, they developed an artificially intelligent solution with superhuman senses that can identify plastic types with near-perfect accuracy.

Because the different plastics are visually indistinguishable, the team relied instead on hyperspectral imaging in the range of 950 to 1,730 nanometers. This technique brings to light otherwise invisible chemical signatures. But interpreting the meaning of those signatures poses a problem. To move past this issue, the researchers turned to machine learning.

A data processing pipeline was developed that began with a principal components analysis to extract the most informative features from the hyperspectral imaging data. Next, a partial least square discriminant analysis was performed to classify the images. This algorithm was trained with a dataset designed to help it discriminate between compostable plastics (PLA, PBAT) and conventional plastics (PP, PET, and LDPE).

During evaluation of the method, it was found that classification accuracy for PP, PET, and PLA had reached 100%. The accuracy for LDPE and PBAT classification was similarly quite good at 90%. The team noted that some of the misclassification errors resulted from the roughness of the material’s surface and scattering of light. Additional work may be required in the future to work around this problem.

At present, work is being done to speed up the rate of the classifications. Before industrial recyclers can adopt these methods, they will need to match the speed of the conveyor systems that are already in place.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles