Ecobot

Image surfaced from the Internet
Loss and mAP during training

Inspiration

The ocean is heavily threatened by plastic pollution. 40% of the ocean surface is polluted by plastic, over a million animals die each year from plastic pollution, 1 in 3 fish caught for human consumption contain plastic, and coral reefs odds of dying increase after encountering plastic. To solve this issue, we have developed Ecobot.

What it does

Ecobot is a YOLOv4 AI model that can detect ocean plastic and locate it with a bounding box.

How we built it

We followed this tutorial for training object detection models. We also used a Bootstrap template for the frontend.

Challenges we ran into

To train the AI, the colab window had to stay open. One of our team members had to wake up at 3 A.M. and 5 A.M. to make sure our AI was still training and that it hadn't ran into any errors. In addition, we had difficulties uploading to github due to the large size of our zip (800 MB). At times, it was also difficult to collaborate due to outside activities and our daily schedules.

Accomplishments that we're proud of

We achieved up to 99% confidence in some bounding boxes! It was also our first time creating an AI model on our own, so we're happy we ended up with a finished product. In addition, most of us had little experience with HTML/CSS, so we had to self-teach ourselves some of the necessary concepts within 48 hours.

What we learned

We learned new aspects of HTML/CSS in order for our website to function more smoothly. In addition, we followed tutorials to create YOLOv4 models for object identification. To connect this all together, we had to work around the clock with frontend and backend development using Flask.

What's next for Ecobot

We hope people implement EcoBot as needed in their projects, so they can have experience understanding how AI Learning works. We also hope that EcoBot can be implemented in an Autonomous Underwater Vehicle (AUV) that could clean up our oceans.