Utilizing AI to Remove Oceanic Plastic Waste-Part 1

Recent papers suggest a way to detect oceanic plastic waste from satellite imagery(image source: NASA)

According to Natural Geographic, it is predicted that the amount of plastic waste in the ocean will reach 29 million metric tons by 2040. This can have devastating effects on the marine ecosystem and our health as much of the plastic is consumed by sea creatures and some of the sea creatures are captured and consumed by humans, plastic still remaining in its bodies. To improve this grim situation, many environmental agencies are searching for a way to efficiently and safely remove plastic waste from ocean waters.

Plastic Litter Project

One of the groups trying to remove plastic waste is the Marine Remote Sensing Group of the University of Aegean. The goal of the group is to find a method of detecting marine pollution from aerial and satellite images. One annual research experiment the group conducts is the Plastic Litter Project. The purpose of the experiment is to develop a way of detecting plastic litter on the sea surface from UAV images taken from aerial drones or satellites. In the experiment targets filled with used plastic water bottles are set afloat on the sea under a controlled environment(so they don’t float away to the open sea). Then, the UAV images of coastal areas where the targets were located were taken from aerial drones and the Sentinel-2 satellite. From the UAV images, the research team finds and recognizes where the plastic targets are located. Each year, the results of the experiment are published as a research paper.

How FDI was created&How it works

In 2018, Lauren Biermann, an Earth observation scientist at Plymouth Marine Laboratory, read the publication of Plastic Litter Project 2018. Using the data collected from Plastic Litter Project 2018 and 2019, Lauren Biermann and her team devised a new method of finding plastic waste on the Ocean called Floating Debris index.

Floating Debris index uses spectral analysis to determine which pixels of a satellite image of the ocean are likely to contain floating debris. Spectral analysis is a technique of analysis that uses the specific wavelengths of light that certain objects(such as clear seawater, floating plastic, or floating algae) efficiently reflect or absorb. For example, clear water absorbs light that is near-infrared wavelengths(about 833 nm). Using this information about which frequencies objects reflect well, it is possible to know which pixels of the satellite image contain objects such as plastic waste, seafoam, or algae.

Based on this method, FDI was created.

The formula for FDI is as follows.

Biermann and her team found four coastal sites(Accra of Ghana, Da Nang of Vietnam, Gulf Islands of Canada, and Scotland of UK) where great amounts of floating plastic waste were reported through social media and applied FDI to satellite images of the four coastal sites. When FDI was applied to the images, several patches of bright pixels(pixels with high FDI values) were detected. While some of the pixels were confirmed to have contained plastic wastes, many of the bright pixels were of other objects such as floating algae or spume.

While FDI was successful in detecting floating material when applied to real satellite images, the objects detected using FDI still needed to be classified in order to fulfill the original purpose of finding floating plastic in the sea. To do this, NDVI(Normalized Difference Vegetation Index) was used alongside FDI to classify floating objects. NDVI is a vegetation index that is based on the notion that an area with healthy vegetation will reflect more near-infrared light compared to red light. This is because the chlorophyll inside a healthy plant would absorb more red light than near-infrared light. To learn more about NDVI, I recommend reading this website.

In the graph on the right side of the above image, you can see that objects of the same kind are clustered near each other when their FDI and NDVI values are graphed.

Finally, to automate the process of detecting sea plastic from satellite images, Biermann and her team trained a Naïve Bayes classification model from FDI and NDVI data of detected pixels. The model achieved an accuracy of 86% in detecting plastics from the FDI and NDVI values.

Thank you for reading this post! In the next post, we will explore the technical side of this method(more specifically the Naïve Bayes method) and implement FDI in python, as well as look at current methods to remove the oceanic garbage that is detected.

Earth image source: https://www.nasa.gov/feature/amazing-earth-satellite-images-from-2019/

Marine Remote Sensing Group official website: https://mrsg.aegean.gr/

All information about FDI was taken from the research paper Biermann, Luren “Finding Plastic Patches in Coastal Waters using Optical Satellite Data”, Nature, 2020.

Andrew Chang is a Student Ambassador in the Inspirit AI Student Ambassadors Program. Inspirit AI is a pre-collegiate enrichment program that exposes curious high school students globally to AI through live online classes. Learn more at https://www.inspiritai.com/.

High school student interested in AI and Programming

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store