Recycling classifier
Which bin does this go in? One photo and an answer
What I want to solve
Almost every child has a phone or tablet at home, and pointing at an object and seeing the answer is a way to learn recycling through play. This project builds that model with a no-code tool, to understand how it is trained and where it fails.
How I build it
- Define the classes. In Spain, four: yellow bin (plastic packaging, cans and cartons), blue (paper and cardboard), green (glass) and general waste. Rules vary by municipality, so this is a starting point.
- Gather data. About 50 photos per class using objects from my home, varying background, light and angle.
- Train and test the model with new objects I haven't photographed before.
- Note where it fails and why.
What I want to check
- How much it depends on the data. If every photo has the same background, what is the model really learning?
- Which cases confuse it. A carton looks like cardboard, but it goes in the yellow bin.
- Which mistakes cost more. Putting something non-recyclable in the yellow bin can spoil the recycling of everything else, whereas leaving a recyclable package in general waste simply loses it. It's the design lesson from the machine learning entry, with much less risk.
What I use
Teachable Machine, a free Google tool that trains image classification models in the browser, no coding needed. It corresponds to the Custom AI level in the machine learning entry.
Results
I'll publish them here once the tests are done: how many photos I used per class, the accuracy per class, the cases that fail and what I learned.
It's an educational tool meant to be used with an adult nearby. It doesn't replace each municipality's official guidance.