RURecycling

RuRecycling

Inspiration

The inspiration for this project came from the fact that not a lot of people know how to recycle properly. When designing our hack, we decided that we wanted to create a software that would be able to visually label an item as recyclable or trash.

What it does

Using AI technology and machine learning, our software will have a camera that will scan the item placed in front of it and determine whether the person should recycle the item or put it in the trash.

How we built it

Using YoloV5 and PyTorch, which are object detection machine learning tools, we imported labeled images into the model which is then trained to differentiate between what we labeled as trash vs recyclable. We gathered data from Google data sheets and used RoboFlow as an assistive tool for labeling to improve the efficiency of our labeling. Within RoboFlow we needed to annotate what is considered trash and what's considered recycling and this required extensive research to make sure that we knew what could be recycled in the state of New Jersey, specifically Middlesex County.

Challenges we ran into

One of the biggest challenges we ran into was finding a large enough data set and making sure that each image was annotated appropriately. This required a lot of research on our end to make sure that we were correctly saying what could and couldn't be recycled even if they're under the same category. For example not all plastics are recyclable; only type one and type two plastics can be recycled, everything else has to be thrown in the trash so when annotating we couldn't just label all plastics and say they were recyclable, we had to make the distinction. Another challenge we ran into was the time it takes for the AI to learn. Due to the large amount of data we needed our AI to differentiate between, it took hours for the AI to be properly trained.

Accomplishments that we're proud of

We are proud and of the amount of data we were able to annotate in a few hours. For AI to get the best reading possible, it needs as much data as possible from a variety of different angles. Having to manually go in and make sure every annotation is correct from our uploaded data sets was a long process but overall it made our AI meticulous in differentiating between our two categories.

What we learned

Using the RoboFlow software was an important skill to learn and was immensely helpful in accomplishing our goal. Prior to this hackathon we knew very little about how RoboFlow worked but now we have a much better understanding of it's usages for machine learning.

What's next for ReduceReuseRU

There are so many more things to teach the general public about recycling. An item can be recyclable but needs to be treated in a certain way so as not to contaminate other recyclable items in a recycling bin. There is also more to discarded items than just recycling and trash such as items considered to be compost. Overall we would like to expand our AI to teach the general public how to treat their recyclables via an informational message provided to users after scanning their items. We would also like to expand our AI to detect composite materials.