Deep learning trained machine vision for object detection and quantification in recycling operations

Narrative

Recycling sectors include scrap metal, construction and demolition (C&D), and household waste. These sectors are multi-billion-pound businesses and have grown drastically in the last couple of decades. Scrap metal recycling alone had turnovers of $217 bn in 2020 globally and is expected to grow to $368.7bn in 2030. While existing recycling operations do recover vast amounts of materials and divert more from landfill, recycling rates in general are still low. The MRF (material recovery facilities) industry typically has recycling rates well below 50% in the UK, largely due to manual sorting and conventional operations. There are also many health and fire risks associated with recycling operations. Lithium-ion batteries, for instance, which are often discarded in household waste, can easily cause fire and explosions during transportation and segregation processes. There is therefore a huge demand and market for new technology in the detection of these hazardous materials.

Prof. Yin at the University of Manchester has teamed up with EMR, on a Proof-of-Concept project in 2023, to explore the feasibility of using AI and vision means for identifying valuable scrap metal pieces in their recycling streams. Cutting edge AI-based technology can not only help detect and recover more valuable materials, but can also help safe-guide the recycling operations at various stages by detecting hazardous materials and objects, significantly impacting sustainable development and the circular economy. Such technology can help modernise and automate recycling processes and increase recovery and recycling rates, while simultaneously reducing contaminations and operational risks. One such example is a prototype AI-vision system developed with Bensons Components Ltd via an Innovate UK KTP project. It has been installed on several recyclers’ sites including a Biffa’s site (since Feb 2021). Since then, both detection performance and false positive rate have been significantly improved.