Hunters

Universal Event | Leveraging AI/ML for plastic marine debris

Marine debris tracking using ECCO-2 with data collected from Twitter and Web App

High-Level Project Summary

We base our project on using two types of approaches to solve the problem. First, we use the application that we developed to allow the general population to send a picture of debris from rivers. The application will ask for the user's approval to access their location using GPS. Afterwards, we use machine learning to detect what type of debris it is. Second, we use twitter API to scrape tweets and use our designed NLP (Natural Language Processing) to detect the mood of the tweet. If the mood is negative, we will use the AI of Google Maps to locate the geographical position and store it in our database. We can view all the data collected on a map which will be based on our database.

Link to Project "Demo"

https://docs.google.com/presentation/d/1JOtM6mPItUt146RN0ufO9eMPWYiLNgoLB-tKpq_9qVc/edit?usp=sharing

Link to Final Project

https://debris-tracker.herokuapp.com

Detailed Project Description

Solving the issue of marine debris is one of the hardest problems Earth is encountering currently. Most of it is microplastics, i.e. plastic that involves a radius smaller than 5mm. Because of the water and the Sunrays, after some time the microplastics become more dense than water, therefore most of it is under the ocean surface. This makes it impossible to collect marine debris and also makes it very hard to track its geographical position with satellite footage.

However, there is a way. We don’t indeed need to find where the debris is in the ocean. We only have to find its sources and then project its trajectory. This is made feasible with NASA’s open source data on global sea surface currents and temperatures. (https://svs.gsfc.nasa.gov/cgi-bin/details.cgi?aid=3912). One research using Estimating the Circulation and Climate of the Ocean, Phase II (ECCO2) has already been done. In that research they distributed particles (that are representing the microplastics) evenly around the world. Our approach aims to find locations of waste that are representing the source of the particles and thus make it more accurate.

We made a web app for people to upload images of debris when they find it to our database. The web app will then classify waste and suggest its prediction. The user can correct our classifier machine learning model. Why is this improvement over the Marine debris tracker app that can be downloaded from google play store? With every image uploaded, we can improve our ML model for trash recognition. Plan for the future would be to put cameras under bridges which would recognize the trash that is passing through and input the location into the database. However, there are downsides to this solution. This project requires a lot of resources and time. Firstly, the marine debris tracker app has only 50k downloads. We would need the government’s help to scale this app and also funds for advertising. This is the most accurate long-term solution.

URL for Webapp: https://debris-tracker.herokuapp.com

While the first solution could give us the most accurate and detailed data, the problem of marine debris is dire. We asked ourselves: instead of collecting data, can we use data that already exists elsewhere? The answer was yes! One of the vastest social networks, Twitter (1B+ downloads and 500 million tweets a day) has a friendly API which we applied to scrape tweets related to debris/waste/trash/plastic and ocean/sea/river/beach (with the AI in their search engine). After that we filtered the tweets further, doing sentiment analysis on the content of tweets which gave us the attitude of the author of that tweet. If the attitude was negative, the author is probably talking about debris in the context of pollution. Finally, we put the tweet into the Google maps search engine (which uses AI that can work out the location mentioned in a tweet), for that we used Google maps search engine. Everything was done in python (to use the code, you have to implement access keys for both Twitter and Google Maps APIs)

Github link: https://github.com/withoutJ/nasa_spaceapps_challange

Space Agency Data

National Oceanic and Atmospheric Administration (NOAA) have conducted an experiment called the Marine Debris: Garbage Patch Experiment. In the experiment, they use particles, which are modelled as the marine debris, and release them in large amounts into the ocean bodies, in addition to using NASA's computerized model of ocean currents called ECCO-2 to distribute them evenly around the ocean.

Most marine debris are microplastics, which are very hard to detect using satellite footages or other such similar sources. So instead, we can track down the path of the debris. Using our collected data set from Twitter, we can get the approximate location of debris and release model particles from those locations to see where the ocean current carries them. Using ECCO-2 ocean circulation model, we will be able to predict the trajectory of the particles.

https://sos.noaa.gov/catalog/datasets/marine-debris-garbage-patch-experiment-drifters-and-model/

https://svs.gsfc.nasa.gov/cgi-bin/details.cgi?aid=3912

https://www.ecco-group.org/

Hackathon Journey

Experience: The Space Apps challenge allowed our team to widen our horizon of knowledge beyond our field of study. There is indeed a difference between academic work and actually putting our thoughts into action and taking steps to solve real-life problems. The feeling of really stepping in and trying to come up with a real-world solution is indeed remarkable.

What did you learn: While constantly searching and deliberating for solutions to our problem, we considered and analyzed complicated solutions to a simple problem. Subsequently, we realized that in this world, full of information and technology being used to solve unique and specialized problems, we can really manipulate and tweak this same information with the simple but groundbreaking technology to solve many other real-world problems. So what we really learned is we just need to think “out of the box” rather than just “work hard”.

Inspiration: We felt that this is an issue which particularly required an immediate attention over any other topic that we came across. Besides trying to make Mars our second home, our team felt we do indeed need to fix our own home first; the home which started our civilization. From that motivation, we resolved to work out solutions that would contribute to making this Earth a bit better place to live in.

Approach developing the project: Our first approach was that we contemplated taking pictures of the rivers (which are the primary source of debris in the first place), using image capturing. Afterwards, we would use object recognition and Machine Learning to identify the objects in the image. Later on our journey of in-depth research, we discovered that there are over 3.5 million miles of river length in the USA alone. Furthermore, according to Google Play, there are only 50 thousand downloads of the app “Marine Debris Tracker”, so this is indeed a possible solution but would need time and resources to be put into practice.

Subsequently, we came up with another solution to the problem, which can be implemented within a shorter period of time than the previous idea of object recognition. We collect information from posts made on social media related to debris being present in seas / rivers and use state of the art NLP (Natural Language Processing) on this information to identify the mood of the post, type of debris and location. Afterwards, we would pin all the locations retrieved on a map.

Overcoming setbacks and challenges: At the beginning, it was indeed a challenge for us to come up with solutions that can be implemented in real life, and we were failing to do so. But as they say, “Team work makes dream work”; we, each of us, tried to come up with our own suggestions, discussed it with each other and as time passed, we had solutions to every issue we had encountered previously.

First, we would love to thank NASA for the opportunity to showcase the work of our team in front of the world. If this challenge wasn’t there, we would not have stepped out of our comfort zone and dealt with the problems of the real-world scenarios. We, each of us, would also want to acknowledge our fellow teammates who have demonstrated their utmost perseverance and dedication to the project's completion.

References