IMPORTANT

Please turn on mixed content to avoid errors when viewing the project https://docs.adobe.com/content/help/en/target/using/experiences/vec/troubleshoot-composer/mixed-content.html

The Creators

Danny Chen Github, Plano West Senior High, Discord Name: Danny Chen
Jimmy Ding Github, Plano West Senior High, Discord Name: Jimmy Ding
Lisa Wu, Plano West Senior High, Discord Name: Lisa Wu

What is Earth Buddy?

Earth Buddy is an app that utilizes deep learning to detect certain objects to tag and recognize whether or not they are recyclable. This data is then fed into a system that gives information on whether they clicked recycle or reuse. For the former, our program provides infographic with concise and brief data pertaining to the proper disposal and the like and finds a proper recycling center if possible. For the latter we gave information sourced from pinterest on how to creatively reuse the items in a fashion that is useful and fun.

So, how exactly does it work?

By utilizing resnet 34’s convoluted neural network we fed the deep learning algorithm a labeled (cardboard, glass, metal, paper, plastic, trash) dataset of around 3000, 512 x 384 images based on an open source dataset. The data was then cleaned and resized in order to help with the accuracy for recognizing certain objects. The data was then trained for a total of around 15 hours using 20 epochs each of 45 minutes to achieve an wonderlous 92.6% accuracy rate with a low loss and validation error using pytorch. According to the desired selection chosen by the user (recycle or reuse), the app gives relevant information using an infographic and utilizes the user’s geolocation in order to find the closest recycling location pertaining to the recyclable object [1]. The latter gives do-it-yourself information scraped from pinterest on how to creatively reuse the object and create a useful “hack”

[1] - https://search.earth911.com/

Why is this important?

With global warming on the rise, recycling has become more of an important focus for the world as environmentalists try to save the planet from further damages. Countries around the globe are setting up recycling systems, but for areas without the convenience of recycling bins, they are not faced with the option to recycle on a daily basis. An american worker can on average produce 1.5 tons of waste per year, having technology that allows people to find out the most convenient way to recycle for them or how they can reuse their products is crucial to reducing the amount of waste we produce. As highschoolers, we see how much paper is wasted at our school because there are no actual recycling systems in our school district. Because of that, we want to find solutions that can let us reuse paper as often as possible or collect our leftover papers from school to recycle elsewhere.

Earth Buddy’s Workflow

We first brainstormed an idea for 3 hours on a current real world problem that we can solve using information we have gathered about us. We based our project off an open source database and trained our data accordingly in pytorch. Then while one developed and hand drew the user interface for the website, another worked on creating the infographics for the appropriate material using canva. We originally planned to hard code diy ideas into the project but we were able to utilize a web scraper for pinterest in order to get creative solutions to reusing it. Finally, we utilized a geolocation api in order to track the user’s location and locate a nearby recycling center for the object using earth911’s database.

Using Earth Buddy

The user, in this case, a homeowner, will need to take a picture of an object they plan on throwing away. Then, immediately after taking a picture, Earth Buddy will use the image and classify whether it is a cardboard, glass, metal, paper, plastic, or trash category. After classifying it into a category, the website tells the user facts about it, interesting DIY projects with the object, and the closest location to dispose of that object.

Additional Functionality

In the future, we hope to add batteries also as a category to furthermore detect whether they exist inside the image and locations to dispose of them. This elevates the project to another level since batteries are trickier to dispose, yet still require special attention.

In addition, even though cardboard can be recycled, pizza boxes with food debris on them cannot. Therefore, we can improve the system to detect with more details and further classify the differences.

Moreover, we want to give the user the option to either retake the image or upload an image from their gallery. This allows for easier use of the app.

What we learned

Since we are all highschool students with limited time and knowledge to implement this project, there are still many things we want to improve on. We learned how to work together as a team even in stressful situations when we all broke down. There were also many front end development bugs we ran into that took some rabbit holes down stackoverflow to resolve. With machine learning, we had to train the model multiple times to find the learning rate as it was not optimized.