From Italy, living in New York. Outside of tech, I love languages and sports.
Github repo:
Project description:
The last century has seen a significant rise in population and production levels, which has
led to an increase in environmental impact and waste. Although global awareness of environmental
impact has been increasing in the last decade, actions of limited impact have caused effective
recycling rates to remain stagnant. According to the Environmental Protection Agency (EPA), the
difference in 2014 and 2015 effective recycling rates was less than one percent. Also, just 35%
of the total waste production in the US was recycled or composted in 2015.
We attribute the slowdown in recycling to the fact that recycling is an effortful and timeconsuming task. This assumption is based on the results of a survey we conducted with 125
responses to people of all ages across the globe. Most people do not take the time to properly
segment their trash into specific recycling bins. For example, in the survey more
than 46% of respondents said they have thrown trash into the landfill bin even if the item could be
recycled. On the other hand, even if people want to recycle, the appropriate bin can be confusing
and lead to contaminated bins and ineffective recycling. The question comes naturally: can we
improve our current recycling rates using machine learning?
Implementation and results:
We designed a computer vision model capable of identifying the material a certain piece of trash is composed of.
Our best model gives us an accuracy of 91.3% on new observations. Compared with the
actual recycling rates in the US (35%), the high accuracy reveals promising results and potential
for adaptability. This model could be implemented into an automatic recycling bin that
recognizes, classifies, and separates trash into different types of recycling bins instantly. Using
computer vision, the bin would be able to recognize if the trash is recyclable or not and store it
accordingly. A person could throw trash away in the same fashion as today, and the bin would do
the rest. If a successful waste sorting occurs in the beginning of the recycling chain, the assumption (strong) there would be
less contamination and recycling facilities would able to drastically increase their recycling rates,
leading to a significantly lower environmental impact. According to EPA, 52.5%
of the total 2015 waste production in the US went into landfill. With our solution, this number has
the potential to be significantly decreased.