Skip to content Skip to navigation

Q&A: Vessel Tracking with Machine Learning

July 15, 2020
Photo by Planet Labs Inc.

Richard Correro is an undergraduate student at Stanford, majoring in mathematics and computational science. He is working this summer with COS to design machine learning systems to analyze satellite imagery of the Earth’s oceans toward combatting illegal fishing.

This work is part of Stanford’s Data Science Institute, which advances data science methods and tools to respond to our most pressing societal and scientific challenges, and Stanford’s Undergraduate Summer Research in Statistics, which provides an opportunity to undergraduate students to engage in interdisciplinary research using statistical methods. The project is a collaboration between COS staff (Shin Nakayama, Fiorenza Micheli, Elizabeth Selig, Colette Wabnitz, and Jim Leape), Trevor Hastie (Statistics) and Serena Yeung (Biomedical Data Science, Computer Science and Electrical Engineering).  


What project are you working on with COS this summer?
Fish is a staple of many people’s diets around the world, and global demand is rising. Although governments and other organizations have undertaken efforts to better regulate fishing activity, there is still a considerable degree of illegal fishing. This leads to pressures on fisheries globally, threatening their sustainability.

Effective regulation and enforcement of laws intended to prevent overfishing and illegal fishing requires good data. There are a number of systems in place to support the detection of illicit activities, including vessel monitoring systems (VMS) and automatic identification systems (AIS), but most of these systems have coverage gaps and not all vessels are mandated to be equipped with these systems. And even vessels equipped with the appropriate technology may be engaged in unlawful behavior. This makes it hard to determine how well regulations designed to ensure the sustainability of fisheries are being followed.

Because of this, interest in using passive observation systems, such as satellites, for observing vessels at sea has been increasing. Access to satellite imagery is now widespread and the capabilities of the satellite platforms themselves have been improving. Platforms like Planet Labs, Inc. now exist which image all of the nearshore ocean globally every single day.

My job is to figure out how to best leverage these new data sources to better understand the behavior and dynamics of fishing fleets. I’m designing machine learning models to automatically extract information from satellite imagery of the ocean, including vessel locations, sizes, etc. Using this new data, I will begin to analyze the dynamics of vessels in Peru’s waters, with the goal of identifying illegal or otherwise nefarious behavior. Hopefully this analysis will be useful for policymakers who wish to design better regulations and enforcement mechanisms.

How did you become interested in this work?
My primary interest is in analyzing the dynamics of human systems using emerging technologies such as machine learning. By shedding light on the behavior of our systems, we can bring attention to problems and support the design of solutions to help to enforce regulations designed to rectify them. To me, these new technologies are immensely powerful, and this power can be used for good – if those who use them decide to do so. Environmental conservation is a cause which is dear to my heart, and the idea that these new technologies may be used to not only understand how humans interact with the environment but also enforce regulations designed to ensure the sustainability of the earth’s natural resources is very exciting to me. That’s why I chose to devote myself to this work.

Why is it important to be designing machine learning systems that analyze satellite imagery of the Earth's oceans right now?
The capabilities of both satellite imagery platforms and machine learning systems are expanding rapidly at the moment. New satellite constellations mean that global coverage is now much greater than it’s ever been, and thanks to rapid advancements in the field of machine learning the kinds of analyses which may be performed on this imagery are now highly sophisticated. Because of the present lack of knowledge regarding our oceans, it seems only natural to use these new technologies to better understand them and how people interact with them.

What's the most interesting part of your work so far? Have you uncovered anything surprising?
With the ever-increasing quantity of data being collected about us and our world it often appears as if all aspects of modern life are well-documented. What I’ve realized in my work so far is that this is only true in certain parts of the world. In other parts, where the financial incentive isn’t there and the capacity to collect tons of data is non-existent, this simply isn’t true. There’s still a lot to be discovered about our planet.

How do you think partners/collaborators will be able to use this work?
The machine learning systems I’m developing may be used anywhere in the world, and I intend to make them freely and openly available. Although I’m focusing on the region contained within the Peruvian Exclusive Economic Zone, I hope that future researchers can use these systems to study other areas so that we can better understand all of the earth’s oceans.

What impact do you hope it will have?
I hope that our research will help to identify behavior which is harmful to our oceans, and that this information will be used by policy makers and experts to design better, more effective regulation. I also hope that the tools I’m working on will be used to better enforce regulations which ensure the sustainability of our fisheries and the oceans themselves.


The two images below show a region near the Port of Lima, in Peru. The highlighted area contains several vessels. 

The two images below show the same area highlighted in red in the above frames. The image on the left is the satellite imagery from highlighted region, and the image on the right shows the vessels labeled by length and width using the machine learning algorithm. 

 

Contact Information

Nicole Kravec
Communications Manager
nkravec@stanford.edu