Open Data Sources for Post-Consumer Plastic Sorting: What We Have and What We Still Need

Abstract

The global plastic crisis is a significant concern. Addressing it requires a multi-faceted approach involving legislative, social, and technological measures to reduce post-consumer plastic waste and increase the recycling rate and material circularity. The EU Packaging Directive mandates a 50% recycling rate for plastics by the end of 2025 and 55% by 2030. Machine learning techniques are increasingly being explored to improve the efficiency of waste sorting, thereby contributing to achieving the aforementioned recycling rate goals. However, their effectiveness relies on the availability of high-quality and comprehensive datasets. One of the challenges in this field is the need for suitable datasets for training and testing machine learning models. Furthermore, access to detailed and real-world data, such as information from recycling plants, is often restricted due to paywalls, limiting the accessibility of critical data for research and development. This review sheds light on the situation of RGB (red, green, blue) image datasets and NIR (near-infrared) data availability to identify current dataset gaps and advocate for open access to relevant data sources. The target is to promote collaboration and innovation to combat the plastic crisis and promote sustainable practices collaboratively.