Recycling waste conserves natural resources, saves energy, and increases economic security by tapping domestic materials to manufacture new products. Accordingly, where economically feasible, it makes good sense to reuse discarded items. However, the recycling process can be time consuming and labor intensive with many tasks, such as sorting materials into classes, being done manually. Labor intensive processes are challenging to scale, which will cause further difficulties in the future as volumes of waste are expected to increase.
Liverpool Hope University researchers think they have a solution to this problem in the form of a low-cost (approx. $130 USD) Raspberry Pi powered waste sorter. Composed of a Raspberry Pi 3 and a camera, the device images trash as it passes by on a conveyor. The images are fed into a machine learning model that was trained to recognize paper, glass, plastic, metal and cardboard objects with the help of the TrashNet dataset.
With the locations of various types of refuse determined, the intent is that the system would eventually interface with a robotic arm to physically move items as needed for sorting. In tests, the device was found to have a 92% average accuracy in classifying objects.
The team is currently facing some issues with speed — the model they are running is apparently a bit of a big ask for a Raspberry Pi 3. Fortunately, there are options that could significantly improve performance without breaking the bank. For no additional cost, the Raspberry Pi 3 could be swapped with a more powerful Raspberry Pi 4. If that was found to be insufficient, a great deal more horsepower-on-a-budget can be had with the NVIDIA Jetson line of single-board computers.
With a bit of polish, a device such as this may find itself in recycling centers around the world in the near future, making the recycling process more efficient while also reducing sorting error that leads to recyclables ending up in a landfill.