Ocean Plastic Detector

(#GER Siehe die deutsche Zusammenfassung unten.)

WHAT DO YOU DO AGAINST OCEAN PLASTIC POLLUTION? I'm developing an artificial intelligence solution that detects floating plastic in the sea in freely available satellite images. This will help us monitor pollution, inform the public, and facilitate clean-up operations. My approach already won an award in mid-May 2019! But I need more time to perfect it until it is really dependable. I am starting a campaign to raise funds to offset my cost of living while I am working on this project. As a freelancer, I depend on paying projects and, unfortunately, I cannot afford to work for free on this one for very long.

The Sentinel-2 satellites of the European Space Agency capture all land surfaces from latitude 56° South to 84° North (most of the inhabited land), coastal waters, and all of the Mediterranean Sea every five days. Their data is free for all to use. In addition to red, green and blue, they capture light at ten more wavelengths. How much information hides in these multispectral images is completely unintuitive: none of us can see in near infrared or in short-wave infrared!

At the highest resolution a pixel is 10m by 10m so it doesn't allow me to see plastic flasks. But there is a strong belief that plastic has a signature in these 13 bands that allows one to detect it if its concentration is high enough. There is a Greek research group who puts out floating islands of plastic bottles, bags, fishing nets into the sea temporarily to display them to the satellites. Through their precious work, we know what kind of signal plastic gives in the satellite images.

In mid-May 2019, my submission was awarded second prize in the category Marine environment at the Sentinel Hub Custom Script Contest! Sentinel Hub provides two web services (Playground and EO Browser) to look at this kind of satellite data without registration. My award-winning script has been released there. By clicking on the links for Evaluate and visualise, you can see that the script produces some false positives. Your backing could change this!

I promise to improve my algorithm and to make it freely available for all to use if this campaign succeeds. Think of the opportunities this will afford for environmental organisations globally and for clean-up teams locally! Depending on how much time I can devote to this, I plan to document the process and release the source code for the community to build on it.

My goals are:

1. Improve my data labelling pipeline to make it more efficient and resistant to user error.
2. Label more data.
3. Analyse the multispectral dataset through various visualisations. Separate plastic from non-plastic in some high-dimensional space using feature engineering.
4. Try out several classification methods of machine learning to identify one or a few good detectors.
5. Translate one of these detectors from the programming language that created it, Python, into Javascript. Make this Javascript program available for everyone via the Sentinel Hub platform.
6. Promote the release of my software on the internets, via all social media channels I can access.
7. Document the project and release the documentation. A more targeted round of promotion.
8. Release the Python code I use for data labelling, data analysis, model fitting, and translating my algorithm into Javascript as free software. A more targeted round of promotion.
9. Potentially submit the documentation as a research manuscript to a scientific journal on remote sensing. I don't think I'll get this far (it would require much time and potentially expert collaborators).
Tasks 1-6 might take one month of my time, Tasks 7-8 another month, and Task 9 a further one. The target goal is a guidance for roughly one month of my time (and income tax, health insurance, pension contribution). I would be grateful for even the smallest donations.

Why you should fund this project:

∙ I know what I'm doing. I have got an MSc degree in mathematics, a doctorate from the University of Oxford (Department of Statistics), and several research publications.
∙ I have been studying and practising methods of artificial intelligence and machine learning for the past 2 years.
∙ I have been programming software to process Sentinel-2 satellite images for the past 6 months.
∙ I feel very motivated because this is an important global issue to solve.
∙ The skills that I mobilise have a price in the job market. Buying this kind of work would cost more than how much I hope to raise. This is my sacrifice for the cause.

Why do I not apply for a research grant?

First, it takes several months to receive a yes/no decision. For my professional and personal situation, doing this project right now is best. Second, this project is smaller scale than what research funds normally support. It would be poor time investment to do the full paperwork, to agree hosting with my latest university, and to potentially establish a collaboration, for the sum that the project requires. Third, I have got no research publication in either machine learning or remote sensing, which would make my application risky.

What can go wrong?

∙ This is a research project, its outcome is uncertain. I see strong signs that my approach can work, but it might work only under a specific set of conditions.
∙ This will not reduce the amount of plastic manufactured, the amount thrown away, the amount ending up in the oceans, and it will not remove plastic pollution from the wild. These are our shared responsibilities. However, this detector will allow us to constantly monitor how bad plastic pollution is, to bring the extent of the problem to the public's knowledge, and to direct clean-up efforts!
∙ There is competition in science and some time in the future, other researchers will attempt to do the same. You should support me because I'm working on this now and not some time in the future; and I've made insights that others might miss.
∙ Sentinel-2 data is not actually available for open oceans, only for coastal waters. Consequently, my detector will work only along the coasts and in coastal waters.

Find me on: Twitter | my blog | LinkedIn | my website. Thank you for joining me on this journey!

Images: stock photos from Pixabay. Myself after my research talk at the Isaac Newton Institute, Cambridge, UK (photo by Vincenzo Abete).

WAS MACHEN SIE GEGEN PLASTIKMÜLL IM OZEAN? Ich entwickle eine Lösung mit künstlicher Intelligenz, die im Meer treibendes Plastik auf frei verfügbaren Satellitenbildern erkennt. Dies würde uns helfen, die Verschmutzung zu beobachten, der Bevölkerung das Ausmaß des Problems beizubringen, und die Umweltsanierung zu koordinieren. Meine Methode hat bereits Mitte Mai 2019 einen Preis gewonnen! Aber ich brauche noch mehr Zeit die Methode so weit zu verbessern, bis sie einen echten Nutzen bringt. Ich starte diese Kampagne, um meinen Lebensunterhalt zu decken, während ich an diesem Projekt arbeite. Als Selbständiger, bin ich auf bezahlte Aufträge angewiesen und kann es mir leider nicht leisten, mich über einen längeren Zeitraum ehrenamtlich mit diesem Projekt zu beschäftigen.

Mein Programm gewann den zweiten Preis in der Kategorie Meeresumwelt beim Sentinel Hub Programmierwettbewerb! Sentinel Hub betreibt zwei Webseiten (Playground und EO Browser), auf denen man sich ohne Registrierung solche Satellitenbilder anschauen kann. Mein gekürtes Programm wurde dort veröffentlicht. Wenn Sie auf den Link unter Evaluate and visualise klicken, dann können Sie sehen, dass mein Programm momentan noch einige fehlerhafte Ergebnisse liefert. Das könnte sich durch Ihre Unterstützung ändern!

Ich verspreche, wenn die Kampagne erfolgreich ist, dass ich mein Verfahren verbessere, und ich es für jeden frei verfügbar mache. Stellen Sie sich vor, welche Möglichkeiten dies Naturschutzverbänden weltweit und Umweltschutzaktivisten in den betroffenen Gegenden eröffnet! Abhängig davon, wieviel Zeit ich in das Projekt investieren kann, plane ich die Veröffentlichung der Dokumentation meiner Forschung und meines Codes, damit die Community meine Arbeit weiterentwickeln kann.


Bence Mélykúti
Freiburg im Breisgau

Your easy, powerful, and trusted home for help

  • Easy

    Donate quickly and easily.

  • Powerful

    Send help right to the people and causes you care about.

  • Trusted

    Your donation is protected by the  GoFundMe Giving Guarantee.