How Computer Vision and Machine Learning can be used to Recognise Different Materials to Make Recycling Easier
This project has been developed by the Materials Made Smarter Centre [ ↵ ] at Swansea University [ ↵ ] in collaboration with the Sustain Manufacturing Research Hub [ ↵ ] and Discover Materials [ ↵ ] to demonstrate how Computer Vision and Machine Learning can be used to recognise different objects to help with the sorting of materials for recycling.
In the past the sorting of materials for recycling at a sorting line has been performed by underpaid people who observe the objects as they pass by on a very fast-moving conveyor belt and make fast decisions about which bin to push the objects into. This is a very tiring and unpleasant job, if you have ever tried to watch the nearby scenery go past whilst travelling in a car, bus or train then you can imagine how tiring and damaging to the eyes this task is. Most people do not last long doing this job before suffering ill-health and there is a high turnover of staff.
As the amount of recycling increases, the speed it must be processed at increases, and soon no human will be able to cope.
This is exactly the type of work Machine Learning and Computer Vision can help with. By training a machine to identify the different objects, mechanical or pneumatic (air-jet) pushers can be directed to sort the material much more efficiently, and with less hazard to health.
The project is divided into four sections:
01 Computing on the Edge [ ↵ ]
02 Working with Limited Resources [ ↵ ]
03 Machine Learning and Neural Networks [ ↵ ]
04 Training for Different Materials [ ↵ ]
a glossary of the highlighted technical terms used can be downloaded here [ ⇓ ]
This project also demonstrates that, with modern advances in hardware, the task no longer requires large expensive computer systems and can be performed by modestly low-powered, edge-computing, devices which are cheap enough to enable hundreds of independent smart sensors, close to the action, each performing a specific task.
Quick Shortcuts to project resources:
download the eBooklet [ ⇓ ], a digital copy of the printed project booklet
download the written guides:
00 Setting up the Demonstration Hardware_Guide [ ⇓ ]
01 Getting the System Up and Running_Guide [ ⇓ ]
02 How to Use the Project Controls to Identify Objects_Guide [ ⇓ ]
03 Investigation of the Code_Guide [ ⇓ ]
04 Improving the Performance for New Objects_Guide [ ⇓ ]
playlist of the walkthrough videos [ ↵ ]
download the demonstration risk assessment [ ⇓ ]
link to the the project github repository [ ↵ ], for advanced technical resources
The platform this project is built on is the Seeed Studio reComputer J1010 NVIDIA Jetson Nano 2GB Platform with the Arm Cortex A57 CPU and NVIDIA Maxwell GPU and it has been developed by Dr R. Gibbs and Prof. C. Giannetti based upon the NVIDIA DLI “Getting Started with AI on Jetson Nano” course [ ↵ ].
Professor C. Giannetti would like to acknowledge the support of the EPSRC (EP/V061798/1) in this Materials Made Smarter Project.
Share this post
