Fur Seals On Bogoslof Island – Post 4

Catching a UAS Drone.

As the island is changing so is our science. We are using new technology, uncrewed aerial systems equipped with different types of cameras to help us count seals. The technology has the potential to reduce the number of people needed in the field to count pups and reduce survey associated costs. We also will be able to minimize disturbance to the animals. Using these systems we still have to confront the challenges of weather – high wind, rain and fog.

We were able to test two different camera systems on Bogoslof and St. George Island in the Pribilofs: standard color-image cameras and infrared cameras (detect an object based on the amount of heat it generates).  We are also hoping to field test multi-spectral cameras (detect an object based on the amount of light it reflects) in the near future and did some studies to figure out how to best do this.

By looking through collected imagery captured with different cameras, we can identify and count the number of fur seals present. It’s exciting because we are hoping to compare these counts from aerial images with the pup counts conducted on the ground by scientists to ground truth with what we have been doing for years.

We have already begun working on our next challenge, which is how to sort through the hundreds of thousands of images that these sort of methods generate, in a more timely and efficient way.

For our Steller sea lion work, we have used a crowdsourcing platform, StellerWatch (hosted by Zooinverse) to help us review the millions of images we collect from remote cameras based at several sites in the Aleutian Islands. Crowdsourcing engages members of the public in the scientific process. StellerWatch has been really successful.  We have been able to review over 400,000 images.  But we know we still need more help.

By teaming up with software engineers, we plan to use Artificial Intelligence (AI)/machine learning to review the images for us.  AI can do it more quickly and efficiently than the human eye. To do this, we need to train a computer to be able to detect and identify fur seals against varied substrates like rocks.  We also have to distinguish between fur seals of different sexes and ages. The computer has to be able to identify animals when they are sitting or lying in different positions. 

To teach the computer, we provide it with a test set of images that we have pre-screened and identified all fur seals. We have taken these steps for analyzing Steller sea lion images from aerial surveys, and soon, we’ll have to do the same for fur seals. This will help the computer learn how to identify them for us.

Our work on the island involved capturing images of all the animals on the island with a color camera.  We also had a chance to test out the infrared camera on St. George in the Pribilofs.  

Both camera systems performed really well. But weather challenges limited our ability to get enough flights this trip to make equivalent comparisons between the human estimates and different methods with the UAS systems.