MarineMarauders

Our machine Learning accuracy vs iterations
A picture of our UI design, with auto-detecting images that get displayed to the user

Inspiration

We wanted to choose a sustainable goal related to trash and decided to focus on underwater trash since we think it's barely talked about and it's a pressing issue.

What it does

The machine learning model is able to classify underwater images to either "Clean" or "Dirty" with around 97% accuracy. "Clean" is chosen if the image doesn't have plastic remains in it. "Dirty" is chosen if it does. We also integrated a front end application which makes this project very user-friendly.

How we built it

We used google colab and tensorflow to be able to code the machine learning algorithim. Google colab/tensorflow made our job much easier and it was very effective way of coding the algo. The frontend was built using PyQt6 in python, using modules such as pillow and tkinter to help with file management.

Challenges we ran into

The main issue we faced was finding good image datasets to train and test the model with. The availability of clear and dirty underwater pictures was limited but we found a comprehensive dataset in the end.

Accomplishments that we're proud of

We are proud of how accurate our machine learning model is for detecting trash, and how we were able to create the project we did in such a short amount of time. We are also proud of how simplistic the UI design is, and how we planned for the future with the design of the front end application

What we learned

In our preparation for the hackathon, we honed our skills in utilizing Google applications, specifically TensorFlow, for tasks ranging from dataset exploration to data cleaning. Our proficiency in these tools empowers us to efficiently navigate data challenges, ensuring a solid foundation for innovative and impactful project development during the hackathon.

What's next for MarineMarauders

Due to limited time, we did not have a chance to try our algorithim on a bigger and more complex database. We would have our model differntiate between different kinds of trash such as water bottles, food wrappers, fishing nets, oil spills, etc. We would also implement a submarine type of robot that would use this algorithim.