Detecting River Plastic Pollution with AI & Satellite Imagery

In the last few weeks, I was working on a fun Computer Vision side-project that can detect plastic patches in the most polluted rivers by satellite imagery.

The Problem:

According to The Ocean Cleanup, 1000 rivers are accountable for nearly 80% of global annual riverine plastic emissions, which range between 0.8 – 2.7 million metric tons per year that mostly end up in our oceans.

The Solution:

A Computer Vision-based model that detects plastic patches by satellite imagery before they’re entering our oceans.

That satellite image down below shows the model’s output, the Pasig River in the Philippines that’s crossing Manila and represents one of the most polluted rivers on our planet.

Bottlenecks:

I found way too less available image datasets – a problem that’s very common for many Computer Vision projects.

AND these datasets which are public include most of the time no high-resolution images what makes the workflow pretty frustrating.

What’s next?

I am pretty satisfied with the model’s accuracy (73%) but it could be improved by more qualitative images and it could be used to make predictions about the direction and speed of the floating plastic patches.

The main intention of this project was to show that AI will produce more positive than negative impacts on our society and defines a technology we all can be excited about.

If you have any questions, recommendations, and feedback about my project, please let me know by commenting down below.