AI for a Better World! We are Doing Our Part. Will You?

Reading Time: 5 mins

We, humans, are the most adapted creatures of the 21st century. Moreover, technological advancement has created a whole new civilization that we live in. According to Wikipedia, “civilization is any complex society characterized by urban development, social stratification, and domination over the natural environment, including the domestication of both humans and other organisms.” 

Yet, somewhere these other species are not advancing like us. 

Why? 

We are just so busy developing lifestyles that we end up hurting other creatures directly or indirectly.

According to a United Nations report released on 6^th May 2019, “One million plant and animal species are on the verge of extinction”. Yeah, you read it right.

Penguins are among them. 

Another 2008 WWF study estimated that 50% of the emperor penguins and 75% of the Adelie penguins will likely decline or disappear if global average temperatures rise above pre-industrial levels by just 2 degrees Celsius—a scenario that could be reached in less than 40 years.

Why are the Fluffy Creatures a Threatened Species?

Characteristics of the Penguins make them sensitive to human activities. They are aquatic flightless birds, which makes it difficult for them to escape from introduced predators. Their Southern hemisphere homes are warming rapidly due to global warming.

They spend long periods of time underwater chasing marine food sources, which are diminishing gradually due to overfishing and oil pollution. It puts the sentinels of Antarctica at a larger risk of starvation. 

Penguins have long lives. The average life expectancy of penguins is 15 to 20 years. They lay just one or two eggs each year, and take several months to nurture and raise them. Egg poaching makes it even more difficult to nurse their species. 

In addition, penguins make use of very wide geographical areas in the ocean when they forage. During winter migrations, they are exposed to a variety of threats such as traveling distances for a meal, or not surviving drenching rainstorms or extreme heat. Such situations and characteristics make them very susceptible to natural and man-made disasters.

Note: Check out Microsoft New about our collaboration to save penguins using AI

Why Should We Save Penguins?

Many of you might be thinking about the importance of the existence of penguins? 

Well, penguins also carry nutrients between the ocean and the land, enriching both with their excrement. Guano harvesting, which is a collection of penguin waste matter is used as a fertilizer. Ultimately, healthy populations of penguins mean healthy oceans.

How Did We Help the Sentinels of Antarctica?

Microsoft – AI for Earth initiative brings together several not-for-profit organizations, which are focused on ecological conservation, and technology organizations to partner together and innovate technical solutions leveraging AI, ML, Deep learning, and more to solve specific conservation problems. 

Gramener is one such data science and technology partner. We have done several projects in association with Microsoft. Detector model is a project that was trained to find animals in Camera Trap images. 

Camera traps are placed in forest areas around the world. When an animal passes by, it captures its motion-sensed pictures. Similarly, millions of pictures are clicked, which are classified into multiple species by scientists all around the world.

Penguin Counting model is yet another project, using Zooniverse’s Penguin watch dataset. Zooniverse considered 16 colonies of penguins in Antarctica and they wanted to keep a track of them in each of these colonies to plan the conservation activities around it. 

The main objective of the project was to count the number of penguins from the images captured by camera traps set up in Antarctica. For this, we leveraged Microsoft’s Deep learning ecosystem (Azure platform, DSVMs) and PyTorch to solve the problem of accurately counting the penguins.

Penguin Dataset Information

• The penguin dataset is a collection of images of penguin colonies in Antarctica, which was set up with the purpose of monitoring their changes in population.
• The images are taken by fixed camera traps in 16 different locations, which have been capturing an image per hour for several years.
• The crowdsourced dot annotations were provided in Matlab format with each dot on a penguin represented as an (x, y) coordinate.

Exploring the Dataset

There were several images without clear visuals. These images had no corresponding dot annotations, which we had to ignore.

Original Dataset

• Training set images: 49156
• Validation set images: 8187
• Test set images: 24598

Cleaned up Dataset

• Training set images: 18525
• Validation set images: 3220
• Test set images: 9062

Issues with Automated Penguin Counting

We faced several issues with respect to the data. 

Firstly, Perspective distortion. A penguin closer to the camera appears larger and takes more image space than the ones at the far end. 

Secondly, Occlusion. Several penguins overlap or appear to hide other penguins. 

Thirdly, Camera setup. Some cameras are set up to take wide shots and some to capture specifics. 

Fourthly, Weather. Harsh weather conditions decrease visibility.  

And finally, density difference. Some areas are densely populated and some are coarse.

Modeling and Approach

We used Multi-Column Convolutional Neural Networks (CNNs) for this problem.

Here, a single image is split into 9 patches, i.e. 3×3 metric. For each patch, we calculated the Gaussian density map. Then, we summarized it into a single count to get an estimated count of penguins, later.

What is the Outcome of Our Efforts?

The impact of our efforts largely benefits the researchers with cost-effectiveness, less time-consumption, and less human intervention.

According to Frontiers, “Penguins evolved over 60 million years ago, and yet perhaps the biggest challenge will be for some of them to survive the next 50 years. Most observed population changes have been negative and have happened over the last 60 years. Today, populations of 11 of the 18 penguin species are decreasing (BirdLife International, 2018), especially those with restricted ranges living in temperate areas close to humans.” 

And hence Penguins reside in extreme climatic conditions, it’s humanly not possible to go and count them. That’s when AI models help researchers and organizations to do the job. Also, the time taken to manually count the penguins in a crowded image is more tedious. Automation can make this process simpler and more economical, minimizing human error. 

How Can You Help?

If you are interested in AI and Deep Learning technology, then join hands with us to improve the model accuracy. You can also help us annotate the images to build better datasets for training models. For more information visit, Penguin Watch – Zooniverse.

However, if you are looking for some hands-on experience then pack your bags and join hands with Dyan DeNapoli who is on a lifelong mission to help save the world’s Threatened and Endangered penguins. She is also known as The Penguin Lady. 

Contact Us to Know More

Gramener is actively contributing to Microsoft’s AI for Earth initiative by building innovative AI-driven models to help researchers monitor, identify, and save species. Here’s a small video produced by GroupNine media and Microsoft describing Gramener’s AI for Good work in a nutshell.

Know more about our offerings on Advanced Analytics and easy data consumption. We are available 24X7 to hear you and help you.

Contact us for custom built low code data and AI solutions for your business challenges and check out computer vision solutions built for our clients, including Fortune 500 companies. Book a free demo right now.