Real-Time Automatic Identification of Plastic Waste Streams for Advanced Waste Sorting Systems

Despite the significant recycling potential, a massive generation of plastic waste is observed year after year. One of the causes of this phenomenon is the issue of ineffective waste stream sorting, primarily arising from the uncertainty in the composition of the waste stream. The recycling process cannot be carried out without the proper separation of different types of plastics from the waste stream. Current solutions in the field of automated waste stream identification rely on small-scale datasets that insufficiently reflect real-world conditions. For this reason, the article proposes a real-time identification model based on a CNN (convolutional neural network) and a newly constructed, self-built dataset. The model was evaluated in two stages. The first stage was based on the separated validation dataset, and the second was based on the developed test bench, a replica of the real system. The model was evaluated under laboratory conditions, with a strong emphasis on maximally reflecting real-world conditions. Once included in the sensor fusion, the proposed approach will provide full information on the characteristics of the waste stream, which will ultimately enable the efficient separation of plastic from the mixed stream. Improving this process will significantly support the United Nations’ 2030 Agenda for Sustainable Development.