Marine Litter Detection via Satellites

Background

​

Marine litter is a major ecological threat to marine ecosystems. It can trap and kill marine life, smother its habitat, act as a hazard to navigation, and ultimately lead to loss of biodiversity [1]. Because of plastic decomposition into the microplastic, tiny toxic particles can also be ingested by aquatic life causing the toxic particles move up the food chain, and finally threaten the health of humans.

​

Besides these, it also has negative economic impact in many areas like the loss of income in tourism sector due to polluted beaches and reduced catch in fishing industry because of damaged ecosystems [2]. In fact, the estimation of losses associated with marine pollution sums up to $500–$2500 billion per year!

​

Therefore, cleaning the seas and preventing further pollution of the oceans is extremely important for our planet, our health, and our economy. To enable this, it is essential to know where marine litter is located.

​

However, currently location of marine litter is mainly found via local surveys and on-ground/sea observations. This approach is expensive, and time-consuming, not scalable, and leads to relatively sparse data resulting in only rough estimations of overall amount of marine litter [3].  Machine Learning, and especially Computer Vision can help to improve the current status quo.

​

Therefore, together with a German Data Science Consulting agency, Alexander Thamm, we recently kicked-off a project that will use Computer Vision and satellite imagery to create a global map of marine litter to create transparency on this topic, facilitate research, and enable better coordination of ocean cleaning efforts!

​

​

Project Goal

 

To create an Artificial Intelligence/Machine Learning (AI/ML) infused Computer Vision system that is able to identify plastic floating in the oceans and seas and other marine litter and their locations and sizes from satellite images in (almost) real-time and is accessible for researchers, environmentalists, activists, authorities, and other organizations for no charge.

​

​

Expected Impact

​

This system will allow better strategical decision making and better coordination of ocean cleaning efforts. In addition, the output of the system (e.g., in form of an interactive map) will be used to strengthen public awareness on this topic and to help further research of this problem.

​

​

Project Phases

​

• Phase 1: Research and Proof-of-Concept
  In this phase we will create first MVP ML model based on public data. It will be able to identify the marine litter, its class, and the size of the garbage patches from public satellite images.

• Phase 2: Collecting more data and improving the model
  If required, we will collect, and label additional data based on known marine litter patches from existing structured surveys to re-train our model improve its accuracy. To speed up this step, we plan to use unsupervised or semi-supervised learning methods utilizing results from Phase 1.

• Phase 3: Creating a platform
  We will create an online application/platform that consumes public satellite data in real time, evaluate them with regard to marine litter, and shows the results on an interactive map to the end-user.

• Phase 4: Running the platform
  Our platform will be open for everyone, especially for environmentalists, researchers, NPOs, and governmental organizations that need data on marine litter for their operations (e.g., scientific research, cleaning efforts, public awareness and transparency).

​

​

Current Status

​

Phase 1 is running.

​

Show Case: Link

​

​

Opportunities for Contribution

 

We are currently searching for MI and domain experts who would be ready to contribute to the project as volunteers.

​

We also would be very grateful for any donations in cash or kind to support this initiative.