Ship or iceberg? If you can tell, there's a reward involved
Monday, November 20, 2017, 5:02 PM - Iceberg? Ship? Teams from around the world that come up with the best ways to make that distinction from space can care share in $50,000 worth of prizes.
Check out the images below.
If the labels weren't there, would you be able to tell which showed icebergs and which showed ships on the water?
One comparison from the C-CORE/Statoil competition. Credit: C-CORE
Here's the real trick, though: Even if you could, can you teach the difference to a computer, so it can make that distinction automatically?
This is the focus of a competition, posted on Kaggle.com - a crowdsourcing site where teams participate in various competitions to develop machine learning solutions to a variety of problems - by the St. John's-based Centre for Cold Ocean Resource Engineering (C-CORE) and Statoil, an international energy company headquartered in Stavanger, Norway.
Icebergs are a real problem in the North Atlantic, as they can pose a threat to ships operating in so-called Iceberg Alley, and to oil rigs on the Grand Banks. Observations from aircraft, ships and oil platforms form the basis for detecting icebergs, currently, but satellite data is becoming more important to the process. C-CORE and Statoil are looking for ways of using this data from space to better identify icebergs, and their hope is that posting their challenge on Kaggle will result in truly innovative methods.
As of November 15, they are reporting that over 1,000 teams have signed up for their Iceberg Classifier Challenge.
"The response has been terrific," Desmond Power, C-CORE’s Vice President for Remote Sensing, said in a news release on the company's website. "Unmatched, in fact."
"Statoil's prize purse of $50,000 has certainly helped," Power added. "But similar competitions with larger prizes have attracted less interest. I guess everyone is fascinated by icebergs. And everyone remembers the Titanic. Helping reduce that threat is a big motivator."
The key to all of this is for a computer to be able to look at satellite radar data, gathered from low-Earth orbit by the Sentinel-1 satellite constellation, and pick out specific characteristics that definitively distinguish a ship from an iceberg. Radar data is preferred over camera images, since the radar can deliver consistent data from day to night and back, but it does have its limitations.
According to C-CORE:
Satellite radar works in much the same way as blips on a ship or aircraft radar: it bounces a signal off an object and records the echo, then that data is translated into an image. An object will appear as a bright spot because it reflects more radar energy than its surroundings, but strong echoes can come from anything solid - land, islands and sea ice, as well as icebergs and ships.
Therein lies the challenge.
Humans are very good at picking out differences and learning to spot those differences in , because generally, that's how our brains are wired. Teaching a computer to mimic this ability is not so easy.
One set of comparison images provided by C-CORE in their competition. HH refers to radar signals transmitted and received horizontally, while HV refers to radar signals transmitted horizontally, but received vertically. The above is considered an "easy classification example". Credit: C-CORE
A much more challenging example. The sets are labelled, but how would you tell them apart? How would you tell a computer to tell the difference? Credit: C-CORE
The deadline for the contest is January 16, 2018, and there are three prizes - $25,000 for the best entry, $15,000 for the second best, and $10,000 for the third best.