Courtesy of Krista Ingram

STANDARDS

CCSS: 6.EE.A.2.C, 6.EE.B.5, MP1, MP5, MP6

TEKS: 7.11A

Article Options

Presentation View
Teaching Resources

Reading Level

AI For Seals!

Artificial intelligence is helping scientists recognize the faces of wild seals  

April 2023
By Jenny Morber

Courtesy of Krista Ingram

Biologist Krista Ingram has always loved spotting a seal during summer trips to the Maine coast. “When I was a young girl, a few seals would be an amazing sight for a summer,” she says. But now, she might see hundreds lounging on rocks in a single day!

Ingram wanted to know why there are so many more seals today—but she didn’t want to harm the animals or their environment. “Currently the methods for observing wild populations of animals include tagging the animal, which means you’d have to catch them and disturb them,” says Ingram. Other less intrusive methods are very time consuming or expensive.

Biologist Krista Ingram took summer trips to Maine when she was a little girl. One of her favorite parts was spotting seals. “A few seals would be an amazing sight for a summer,” she says. Now, she might see hundreds lounging on rocks in a single day!

Ingram wanted to know why there are so many more seals today. Counting wild animals often requires tagging them. That means she would have to catch and disturb them. She didn’t want to harm the animals or their environment. Other less intrusive methods are very time consuming or expensive.

Then she read an article where scientists used human facial recognition and artificial intelligence to identify lemurs. Could it work for seals? She had to try. Ingram and her students at Colgate University worked with mathematical biologist Ahmet Ay to develop artificial intelligence (AI) to recognize seal faces. AI is software trained to perform a task by giving it many instances of that task and allowing it to “learn.” They call the program they developed SealNet.

To retrain AI originally designed to identify humans, Ingram needed pictures of seals—a lot of them! She took her students and two children to the Maine coast to “take photograph after photograph of the same seals with their faces in slightly different orientations,” she says.

They started with 50 seals and gave each a name and a number. Then SealNet got to work matching the faces. The team told SealNet when it made a match and when it got it wrong. The database now has 1,024 identified seals, with names like Clove, Petal, and Flash.

Then Ingram read about scientists who used human facial recognition and artificial intelligence to identify lemurs. Could it work for seals? She had to try. Ingram and her students at Colgate University paired up with mathematical biologist Ahmet Ay. They developed artificial intelligence (AI) to recognize seal faces. AI is software trained to perform a task. Scientists help it “learn” by giving it many instances of that task. They call the program they developed SealNet.

Ingram needed a lot of pictures of seals to retrain AI originally designed to identify humans. She took her students and two children to the Maine coast. They took “photograph after photograph of the same seals with their faces in slightly different positions,” she says.

They started with 50 seals. Each got a name and number. Then SealNet got to work matching the faces. The team told SealNet when it made a match and when it got it wrong. The database has 1,024 identified seals.

Kristin Shoemaker via Flickr

Today SealNet can identify seal faces with 88% to 96% accuracy. That’s almost as good as human facial recognition software! The software is free, so any scientist studying seals can use it to track where seals go, see how they interact with each other, estimate their populations, and more.

What was Ingram’s favorite part? That her idea worked! Often, answering questions in conservation biology can take a long time, she says, but SealNet provides quick results. With math, “you can run tests of things, you can try things out really easily,” she says. “There’s a power to that that’s really satisfying.”

Today SealNet can identify seal faces with up to 96% accuracy. That is almost as good as human facial recognition software! Any scientist studying seals can use the free software. It can help them track where seals go, see how the animals interact with each other, estimate their populations, and more.

Ingram’s favorite part was that her idea worked! Answering questions in conservation biology can take a long time, she says. SealNet provides quick results. Researchers “can run tests and try things out really easily” using math, she says. “There’s a power to that that is really satisfying.”

The Mark-Recapture Formula

Before developing SealNet, Ingram and her students used the Mark-Recapture technique to estimate seal populations. In this method, you capture and mark a group of animals and release them back into the wild. Later you catch another group and see how many have the mark. If the population is small, many of the recaptured animals will be marked, but for a large population, only a few will be.

This relationship is described by this formula:

VARIABLES:

N = estimated number of individuals in the population

M = number captured and marked

C = total number captured the second time (with and without a mark)

R = number recaptured (those with a mark)

The Mark-Recapture Formula

Before developing SealNet, Ingram and her students used the Mark-Recapture technique to estimate seal populations. In this method, you capture and mark a group of animals and release them back into the wild. Later you catch another group and see how many have the mark. If the population is small, many of the recaptured animals will be marked, but for a large population, only a few will be.

This relationship is described by this formula:

VARIABLES:

N = estimated number of individuals in the population

M = number captured and marked

C = total number captured the second time (with and without a mark)

R = number recaptured (those with a mark)

USING A FORMULA

Ingram’s team took pictures of seals and used SealNet to identify them. You can use the Mark-Recapture formula to estimate seal populations where the initial photographs are the “capture” and SealNet’s photo identification is the “mark.”

EXAMPLE

Ingram’s team photographed and identified 50 seals. A week later, they “captured” 30 seals on camera, and SealNet recognized 8 from the first set of photographs. What’s the estimated seal population?

STEP 1

Replace the variables in the Mark-Recapture formula with the values you know.

STEP 2

Isolate the unknown variable, if necessary, and solve.

STEP 3

Round your answer to the nearest whole number.

N = 187.5 ≈ 188

So the estimated seal population is 188 seals.

USING A FORMULA

Ingram’s team took pictures of seals and used SealNet to identify them. You can use the Mark-Recapture formula to estimate seal populations where the initial photographs are the “capture” and SealNet’s photo identification is the “mark.”

EXAMPLE

Ingram’s team photographed and identified 50 seals. A week later, they “captured” 30 seals on camera, and SealNet recognized 8 from the first set of photographs. What’s the estimated seal population?

STEP 1

Replace the variables in the Mark-Recapture formula with the values you know.

STEP 2

Isolate the unknown variable, if necessary, and solve.

STEP 3

Round your answer to the nearest whole number.

N = 187.5 ≈ 188

So the estimated seal population is 188 seals.

Use the formula to estimate seal populations. Round answers to the nearest whole number. Record your work and answers on our answer sheet.

Use the formula to estimate seal populations. Round answers to the nearest whole number. Record your work and answers on our answer sheet.

Ingram identified 25 seals at one beach. When she returned, she photographed 11 seals, and SealNet recognized 3 of them. What’s the estimated seal population at that beach?

Ingram identified 25 seals at one beach. When she returned, she photographed 11 seals, and SealNet recognized 3 of them. What’s the estimated seal population at that beach?

A. On another trip, Ingram and her students identified 83 seals. A month later, they returned and photographed 51 seals, of which SealNet recognized 15. What’s the estimated seal population?

A. On another trip, Ingram and her students identified 83 seals. A month later, they returned and photographed 51 seals, of which SealNet recognized 15. What’s the estimated seal population?

B. What would the estimated population on this beach be if SealNet had recognized 26 seals out of the 51 photographed?

B. What would the estimated population on this beach be if SealNet had recognized 26 seals out of the 51 photographed?

If a beach’s seal population is 450 seals and the team identifies 50 seals the first day, how many would the team expect SealNet to recognize of the 31 photographed the next week?

If a beach’s seal population is 450 seals and the team identifies 50 seals the first day, how many would the team expect SealNet to recognize of the 31 photographed the next week?

How does the number of seals SealNet recognizes, or “marks,” affect the the population estimate? Explain your reasoning.

How does the number of seals SealNet recognizes, or “marks,” affect the the population estimate? Explain your reasoning.

Algebra
Equations
Fractions
Fraction Division
Skills Sheets (4)
Skills Sheets (4)
Skills Sheets (4)
Skills Sheets (4)
Games (1)
Lesson Plan (1)
Article (1)
Text-to-Speech