Artificial Intelligence (AI) is currently being used in a number of ways across different industries. From building digital assistants to measuring accessibility to early detection of cancer cells, AI is creating an impact in various fields. One of the areas where you would not expect AI to be used is counting Puffins.
Puffins: Why is it necessary to count them
Counting Puffins is a difficult job. It traditionally requires rangers to lie on the ground and put their hands into burrows to feel for a pair of Puffins and their egg. Puffins don’t like this kind of prodding in the comfort of their own homes and often react by giving the invasive hands a bite or a scratch.
In the process of counting Puffins across tens of thousands of burrows, the rangers go through a lot of plasters. While Puffins are not at imminent risk of extinction, they are placed on the Birds of Conservation Concern 4 Red List, meaning that there are serious concerns over numbers in the wild. This concern is exacerbated by the fact that Puffin couples only lay one egg a year.
As a result, the rangers brave the risk and take part in counting projects in places such as the Farne Islands, the Shetland Islands and the Isle of May, off the east coast of Scotland. The goal of this counting is to understand whether their population is growing or shrinking.
How can AI help count Puffins across thousands of burrows
AI as a technology is going to have a far-reaching impact on humanity and one of its impacts could come in the form of maintaining ecological balance. SSE Renewables is piloting a new way to count Puffins using artificial intelligence, machine learning and image recognition technology on the Isle of May. The project, supported by Microsoft, Avanade and NatureScot, could change the way animal colonies are counted and also highlights participation from building development companies.
SSE Renewables, part of FTSE 100 company SSE, is a leader in renewable energy across the UK and Ireland. The company sees sustainability as a top priority and is working with tech giant Microsoft to identify, develop and deploy innovative solutions aimed at reaching zero-carbon ambitions. The company is also working to “spot, recognise and count puffins using technology”. This could be the first big step in minimising the disruption to the birds’ breeding and feeding habits.
Here is how the technology works. SSE Renewables and Avanade have placed a total of four cameras in stainless steel boxes on the island to capture live footage of the Puffins. This process starts after Puffins begin their return to land to breed in late March or early April following eight months at sea.
Puffins make the island their home for the spring and summer months and often use the same burrows as previous years. Each of the boxes housing the cameras include a condensation heater, wipers to cope with the miserable weather and a backup power supply. The data captured by these cameras is stored in a Microsoft Azure Data Lake.
SSE Renewables also use the Azure Kubernetes Service, a serverless Kubernetes program with power and elasticity to handle huge amounts of information. “The innovative Puffin monitoring project on the Isle of May demonstrates the impact technology can have on advancing sustainability and is just one initial example of how we are collaborating with SSE to shape a more sustainable future,” said Clare Barclay, Chief Executive Officer at Microsoft UK.
Avanade and SSE Renewables have trained artificial intelligence using an “image recognition” model. This model is fed sample pictures until it learns to recognise structures such as lines, squares and circles. The subsequent images fed to the AI model become more complex, allowing the AI model to recognise objects.
Once the AI model is capable of recognising objects, Avanade and SSE Renewables feed the model with images of Puffins and other random images. The model is also taught which images are of Puffins and which are not. Once this training phase is turned and the cameras on the Isle of May are turned on, the AI will be able to spot the Puffins. It will also be able to separate them from background images such as rocks. The cameras are also capable of tracking Puffins frame by frame as they move around their environment.
How does AI recognise each Puffin
The AI will work by drawing a box around each Puffin it spots and giving them unique tags like ‘001, 002, 003’ etc. When the camera moves to the next frame, the AI with its learning data is capable of understanding that the Puffin closest to a particular box is the same Puffin and it has just taken a step to the left or right and it will redraw the box around the bird.
“This happens over and over again for every frame of the footage. Even if the Puffin flies out of the frame, the AI system will recognise where it flew out of sight, and attempt to track it again if it comes back. That’s how we can track and count individual Puffins,” said Simon Turner, CTO of Data and AI at Avanade.
Puffins are mostly nocturnal when they nest on land, meaning they are awake at night and asleep during the day. As a result, the researchers are required to adjust the camera’s brightness and resolution settings to capture footage of the birds. They cannot use night vision cameras since the AI model is trained on images of Puffins captured during daylight. Puffins also have very good vision, which helps them see well at night.
An estimate shows that there are around 80,000 Puffins recorded on the 140-acre island in March 2020, making it the third-largest colony in the UK. This makes the task a bit difficult for AI but also sets a precedent for AI’s use in understanding animal colonies.
With this new AI-based technology, SSE Renewables is not only trying to learn how to count the Puffin population but also learn how its wind farm affects the birds’ flightpath as they travel to gather food to take back to their burrows. SSE Renewables, Microsoft and Avanade have already run this experiment once with Puffins having hatched and raised their pufflings and left the Isle of May for another season.
With the data at their disposal, these companies will learn whether their AI project succeeded in its mission of counting Puffins population without being an intruder. On the basis of the success of this project, the AI model could be extended to monitor other animals who may be affected by developments like wind farms.