AI to count wild animals in Tanzania

You can not think of anything like this with artificial intelligence (AI). From digital assistants to chatbots, the developments are fast. Researchers are now considering a new application of AI.

Wild animals in the picture

Researchers at the University of Wyoming are developing AI that allows wild animals to be identified and counted. Nice piece of technology is behind this. Artificial intelligence is only truly intelligent if at least what people can do. Therefore, the researchers compared the performance of AI with how well people could identify the animals.
3.2 million photographs of wild animals in Tanzania were made with hidden cameras that react to movement. These photos were taken for the project Snapshot Serengeti, where photographs of wild animals in their natural habitat are classified by citizens. Some 50,000 volunteers have tagged the photos of the animals in recent years.

Artificial intelligence versus human

In order to be able to identify the animals with AI, the researchers used machine learning techniques. Neural networks are a reflection of how the human brain sees patterns in the world. The system has been trained with these deep neural networks. The software counted, labeled and described a series of animal photographs. The AI screened 3.2 million animal photos in a number of weeks. The system described which of 48 species it was, how many of these species were present in the picture and what the animals did, such as sleeping, eating or moving. The AI performed almost optimally, with an accuracy of 99.3 percent. And the people? They were 99.6 percent good.

Rangers may not have to sit in their zebras on a rocking jeep with their binoculars in the future to count lions. The AI can come in handy when studying changes in ecology, biology and the behavior of wild animals.


Subscribe to our Newsletter