Sorting garbage into different types ready for recycling is exactly the sort of image-recognition task AIs should be good at, and a team based in South Korea has put together a really clever project that does just that, with Raspberry Pi power, via Hackster (opens in new tab).
Created by Team Techsquare as a means to help citizens in the Republic of Korea to learn how to correctly sort their garbage. Powering the project is a Raspberry Pi 4 (opens in new tab) board and V2 Camera Module. Image recognition speeds aren't fast, to improve them you would need a Tensor Processing Unit (TPU), but they are fast enough for the task at hand. The machine is at a prototype stage, so doesn’t look much, with its wooden frame and copious amounts of tape, but the software is working, which is the crucial part of this green project.
On top of the Raspberry Pi OS there's OpenCV, TensorFlow, and the OneM2M platform. Trash is fed through the wooden frame under the eye of the camera, and the image is analysed on the Pi using a model trained in TensorFlow , sorting it into categories that require different recycling treatment. Should the visual sorting fail, there's also a spectrometer to perform secondary sorting. An actuator then sorts the object into the right can.
If you’re interested in making your own, there are schematics (opens in new tab) and code on the Hackster site.