Save penguins with AI

Penguin populations (such as the most adorable herds of animals) are threatened by environmental disasters, including climate change, but to protect these populations, as in the case of whales, you must first understand them and understand the challenges. has been noticed. Intel and Gramener are using AI to analyze the population of Antarctic penguins and highlight new research that paves the way for better protection of fragile ecosystems.

Save penguins with AI

Gramener, a data science consulting firm, has partnered with Microsoft’s AI for Good program to address these challenges. The company began using data from Zooniverse’s Penguin Watch Tool. The output of the tool consisted of a penguin camera trap image (distributed on about 40 strange pages) manually tagged by a volunteer citizen scientist, showing where the penguins are in the pictures.

Gramener passed this tagged data (about 50,000 images) on to the deep learning model and was equipped with a density-based counting method to estimate the number of penguins in a particular group in the image. We then validated the model with another data set of more than 8,000 images. To train the algorithm, Gramener used a scalable Intel Xeon processor and a virtual machine with Microsoft Azure.

The researchers faced many obstacles. Perspective distortion (penguins near cameras taking up space in the image), overlapping penguins, differences in camera traps, weather conditions, and other factors have made the analysis a serious challenge. Upon completion, the tool was optimized for performance, eventually with training times of days or hours, allowing researchers with limited resources to apply it.

“Today, it’s important to understand the impact on the penguin population in Antarctica on Penguin Awareness Day,” said Naveen Gattu, COO and co-founder of Gramener. “AI is proud that researchers have the power to identify the causes of decline and that they use Intel AI technology to apply their social impact. Our crowd-counting solution has the potential to help you better understand penguin populations.

Gramener hopes it will help researchers avoid counting through the tedious (and often inaccurate) hands of penguin populations, which can delay research and decision-making. Intel also highlighted how the approach Gramener used has applications beyond Antarctica, such as counting people in stores and counting cells with microscopic images. Finally, Gramener commented on the images of the penguin watch and urged interested readers to “put their hands together” to improve the accuracy of the model.